Longitudinal Transitions between Internalizing and Externalizing Symptoms and Associations with Substance Use among US Young Adults, 2016–2023

Abstract

Background:

Internalizing and externalizing symptoms are key mental health indicators that frequently co-occur in young populations. This study examined transition probabilities between none/low-symptom, exclusive internalizing (e.g., depression, anxiety), exclusive externalizing (e.g., aggression, impulsivity), and co-occurring states in young adulthood and assessed associations with substance use.

Methods:

Using four waves (Wave 4–7) of the Population Assessment of Tobacco and Health Study (2016–2023), this prospective cohort study applied Markov multistate transition modeling to estimate transition probabilities across 4 mental health states based on self-reported past 12-month symptoms. We then examined how current (daily, non-daily) nicotine use, past 30-day binge drinking, cannabis use, prescription drug misuse, and sociodemographic characteristics were associated with transitions between mental health states among baseline young adults (18–24 years).

Results:

In the sample (N=5,575), the baseline prevalences of none/low symptom, exclusive internalizing, exclusive externalizing, and co-occurring states were 62.5%, 12.1%, 10.4%, and 13.9%, respectively. Sizable transition probabilities from internalizing or externalizing symptoms to co-occurring symptoms were observed in both short-term (21.6% and 19.7% at Wave 2) and long-term (14.7% and 14.4% at Wave 4) analyses. Co-occurring states demonstrated greater stability (43.9% short-term; 16.4% long-term) than internalizing or externalizing symptom states. Nicotine use was associated with transitions from none/low to internalizing symptoms; binge drinking and prescription drug misuse with transitions to externalizing symptoms; and cannabis use with transitions to both internalizing and externalizing symptoms.

Conclusions:

Transitions to co-occurring symptoms are frequent in young adulthood, and may be associated with patterns of substance use.

INTRODUCTION

Young adulthood, marked by significant psychological, social, and neurobiological changes,^1,2 is also a period of heightened vulnerability to mental health problems.^3,4 In 2022, 23.1% of U.S. adults (59.3 million) reported any past-year mental illness (e.g., mental, behavioral, or emotional disorder), with the highest prevalence among young adults (36.2%).⁵ Past-year serious mental illness (e.g., a mental, behavioral, or emotional condition that causes significant functional impairment and substantially limits major life activities) was also most common in this group (11.6%) compared to other age groups (26–49 years: 7.6%; 50+ years: 3.0%).⁵

Common mental health problems are commonly categorized into two different domains: internalizing and externalizing symptoms.^6–8 Internalizing symptoms broadly refer to inwardly-directed distress⁹ (e.g. depression, anxiety) while externalizing symptoms refer to outwardly- oriented behaviors such as aggression and impulsivity.⁹ While these domains are characterized by distinct developmental trajectories and risk profiles, they are not mutually exclusive and often co-occur, particularly among adolescents and young adults.^10–12 This co-occurrence is concerning because young people with both symptoms may experience greater impairment.¹³

Mental health symptoms in earlier life stages tend to decline with increasing age.^14,15 However, some longitudinal studies indicate that for a subset of individuals, earlier-life mental health symptoms may persist or recur into adulthood.^16,17 Furthermore, young people often transition between different mental health states.¹⁸ For example, internalizing problems in adolescence predicted externalizing problems in young adulthood in a Dutch longitudinal cohort study (baseline mean age = 11 years; final follow-up mean age = 26 years).¹⁹ In a narrative review (without meta-analysis) focused on population-based studies, within-person cross-lagged associations between internalizing and externalizing symptoms were often modest,²⁰ suggesting that the evidence of mutual influence between two domains is inconsistent. Less is known about shifts from single-domain presentations (e.g., only internalizing or only externalizing symptoms) to co-occurring symptom profiles, and vice versa, although the co-occurrence of internalizing and externalizing symptoms has been linked to greater overall symptom burden and higher psychiatric comorbidity.²¹ Thus, transitions into or out of the co-occurring state may reflect clinically meaningful changes in mental health.

Mental health problems and substance use often co-occur. Prior systematic reviews have documented associations between substance use (e.g., nicotine, alcohol, cannabis, opioid) and mental health conditions.^22–27 Some of these studies also suggest that these associations may operate in both directions—mental health problems can precede or precipitate substance use, and substance use can also contribute to subsequent mental health problems. For example, a systematic review of Mendelian randomization studies reported evidence of a bidirectional and causal association between cigarette use and symptoms of mental disorders, including depression, bipolar disorder, and schizophrenia.²² Similarly, other systematic review and meta-analysis studies reported reciprocal associations between mental health problems (e.g., mood disorder,²⁵ internalizing and externalizing symptoms,²⁶ depression²⁷) and substance use outcomes (e.g., tobacco,²⁵ cannabis,²⁵ and alcohol use^26,27), particularly among adolescents and young adults.^26,27 Use of these substances is prevalent among young adults; however, there is limited understanding of whether substance use is associated with increased likelihood of transition into internalizing, externalizing, and co-occurring symptom states, or with decreased likelihood of transitioning into lower-symptom states. In addition, while prior research typically examines incident mental health symptoms using binary follow-up thresholds,^22–27 a transition-based framework can capture movement across multiple symptom states (e.g., recovery, persistence, and progression), providing a more comprehensive understanding of mental health dynamics over time than a simple onset model.

Markov multistate modeling is a mathematical framework that captures transitions among a sequence of states over continuous time using repeated measurements.²⁸ This method offers the analytical advantage of modeling transitions between health states (e.g., escalation in symptom severity or improvement to a less severe state) while efficiently incorporating longitudinal data to estimate transition intensities and probabilities.²⁸ This approach further allows estimation of the relative duration spent in each state (i.e., sojourn time) and the associations between key explanatory variables and transition rates, thereby providing insight into factors that may be associated with the likelihood of transition between states.²⁸ Markov multistate models have been increasingly applied in epidemiologic research.^29–31.

In this nationally representative longitudinal study of U.S. young adults, our primary analyses examined transitions over a 4-wave period (2016–2023) between 4 mutually exclusive mental health states: (1) none/low symptoms, (2) past 12-month exclusive internalizing symptoms, (3) past 12-month exclusive externalizing symptoms, and (4) co-occurrence of internalizing and externalizing symptoms. As a secondary aim, we assessed whether four substance use behaviors, nicotine, alcohol, cannabis use, and prescription drug misuse, were associated with an increased likelihood of adverse mental health transitions.

METHODS

Study Design and Participants

Data were drawn from young adults aged 18–24 in the Population Assessment of Tobacco and Health (PATH) Study, a large, nationally representative, prospective cohort of non-institutionalized individuals in the U.S. with seven waves of data collection. The PATH Study is designed to collect annual data on substance use, behavioral and mental health, and other health-related risk and protective factors. The survey employs a four-stage stratified area probability sampling design. Comprehensive details of the PATH design and methodology have been published elsewhere.³²

This analysis used four waves of the PATH Study (Wave 4: December 2016–January 2018; Wave 5: December 2018–November 2019; Wave 6: March 2021–November 2021; Wave 7: January 2022–April 2023). A replenishment sample was introduced in Wave 4, creating a new cohort and revising study design and estimation procedures;³³ thus, only data from Wave 4 onward were analyzed. Wave 4 served as the baseline wave for our analyses (N=11,283). At Waves 5–7, 2,020, 2,441, and 1,219 participants were lost to follow-up. Of 5,603 eligible participants (22,412 observations), analyses were restricted to those with complete four-wave data. Because the PATH Study reports age in categorical ranges rather than as a continuous variable, mean age could not be calculated; participants who were followed through Wave 7 were approximately 24–30 years old at Wave 7. We excluded 3 individuals with a single observation and cases missing outcomes (374 observations) or weights (98 observations), yielding 5,575 participants, 21,928 observations, and 16,353 transitions (eFigure 1). The WESTAT Institutional Review Board obtained informed consent for participation. This study was not pre-registered.

Measures

Mental health states.

At each wave, internalizing and externalizing symptoms were assessed with the Global Appraisal of Individual Needs–Short Screener (GAIN-SS),³⁴ a validated screener adapted from the full GAIN³⁵ and recommended by the PhenX Toolkit for epidemiological research.³⁶ The PATH Study assessed internalizing mental health problems with four items from the GAIN-SS—depression, psychological distress, anxiety, and sleep difficulties.^37,38 Externalizing mental health problems were assessed using seven items from the GAIN-SS, supplemented with two hyperactivity items from the full GAIN Behavioral Complexity Scale, covering difficulties with attention, difficulties following instructions, deceptive or manipulative behavior to obtain desired outcomes, bullying or threatening others, initiating physical fights, restlessness, and impulsively responding before a question was completed.^37,38 Prior work supports the reliability of PATH internalizing and externalizing measures (Cronbach’s $a=0.72–0.82$)³⁹ and the broader validity of PATH instruments.⁴⁰

For each item, participants were asked when they last experienced the problem, with response options of past month, 2–12 months, >1 year, or never. Derived from the categorizations in prior research,^34,35,39 both internalizing and externalizing symptom measures were dichotomized as none/low/moderate (<4 symptoms in the past 12 months) versus high (≥4 symptoms). We then used these two binary indicators of internalizing and externalizing symptoms to categorize 4 mental health states at each wave: none/low, exclusive internalizing, exclusive externalizing, and co-occurring symptoms.

Substance use.

Participants’ substance use behaviors were measured using the following time-varying (measured in each wave) items: (1) current established nicotine use (current every-day or some-day use of cigarettes and had smoked more than 100 cigarettes in their lifetime, or current every-day or some-day use of e-cigarettes and had used electronic nicotine products fairly regularly), (2) past 30-day binge drinking (4+ drinks for female and 5+ drinks for male respondents on the same occasion in the past 30 days), (3) past 30-day cannabis use (hash, tetrahydrocannabinol [THC], grass, pot, weed or cigar/cigarillo/filtered cigar containing marijuana), and (4) past 30-day prescription drug misuse (stimulants such as Adderall and Ritalin, painkillers, sedatives or tranquilizers). All these variables were coded as dichotomous and analyzed as separate measures. These measures were defined based on prior epidemiologic studies using the PATH Study for each substance.^29,41,42

Statistical Analysis

We applied a Markov multistate transition modeling approach to estimate the transition probabilities between the 4 mental health states we defined earlier. Instantaneous transitions among these states are depicted in Figure 1. Transition probabilities were derived from transition intensities calculated across the 4 study waves where transition intensities represent the instantaneous risk of moving from one state to another.²⁸ Transition intensities, the instantaneous rates (or hazards) of moving from state $r$ to state $s$, were estimated as:

$$q_{rs}\left(t,z\left(t\right)\right)=\underset{\mathit{\delta }t\to 0}{\text{lim}}P\left(S\left(t+\delta t\right)=s|S\left(t\right)=r\right)/\delta t$$

where $t$ denotes time, $z\left(t\right)$ represents covariates at time $t$, and $S\left(t\right)$ indicates the individual’s state at that time. The transition intensities comprise a 4 × 4 matrix Q (reflecting four possible states), with each row summing to zero ($q_{rr}=-{\sum }_{s\ne r}{q}_{rs}$). The diagonal elements ($q_{rr}$) represent the rate of remaining in the same state, while the off-diagonal elements ($q_{rs}$) indicate the instantaneous rate of transitioning from state $r$ to state $s$. Transition intensities were estimated using maximum likelihood methods. The probability that an individual’s next transition is from state $r$ to $s$ is given by $-q_{rs}/q_{rr}$. Transition probabilities (i.e., the likelihood of being in a given state after time $t$) are derived by taking the matrix exponential of the scaled transition intensity matrix, $P\left(t\right)=\mathit{\text{expm}}\left(tQ\right)$. A continuous-time Markov multistate model represents a finite-state stochastic process in which the probability of moving to a future state depends solely on the current state, independent of prior history. In these time-homogeneous models, exact transition times are unobserved, and changes between states are assumed to occur continuously over time, allowing for the possibility of multiple transitions between observation points.

Figure 1.

Direct Transitions Modeled between Mental Health States^a

^a Mental health states were mutually exclusive and assessed based on self-reported symptoms within the past 12 months. Internalizing and externalizing problems were coded as binary indicators: none/low (<4 symptoms in the past 12 months) versus high (≥4 symptoms in the past 12 months). The category ‘co-occurring mental health symptoms’ refers to the co-occurrence of (≥4 symptoms of internalizing and externalizing problems. All directional paths are represented with double-headed arrows, indicating that bidirectional transitions between all mental health states were permitted in the model.

As primary analyses, derived from prior research,^29–31 we estimated mental health transition probabilities in the short term (at 1 wave later from the baseline,i.e., Wave 2) and long term (at 3 waves later from the baseline,i.e., Wave 4) using separate models. We compared the estimates from these models to assess the consistency in transition dynamics over short- versus long-term intervals. We also reported the mean sojourn time (i.e., the expected duration an individual remains in a given state before transitioning), estimated from the transition intensity matrix for the overall study sample.²⁸ We allowed for transitions to a non-adjacent state (e.g., from none/low mental health symptoms directly to co-occurring symptoms, or vice versa).

For the secondary aim, we estimated separate models were estimated for each substance use behavior (nicotine, binge drinking, cannabis, prescription drug), controlling for sociodemographic characteristics (sex, race, ethnicity), to evaluate their associations with transitions across mental health states. Supplementary analyses further examined how sociodemographic characteristics, including sex, race, and ethnicity, were associated with mental health transitions. These associations were presented as hazard ratios (HRs). In the secondary and supplementary analyses, transitions between internalizing and externalizing symptom states and transitions from the none/low-symptom to co-occurring symptom states were constrained to zero because of the limited number of substance use and sociodemographic cases for these transitions, preventing the HRs from approaching infinity within the 95% confidence intervals (CIs).^29,43

This study incorporated weights to account for the complex survey design of the PATH Study, producing nationally representative estimates.⁴⁴ We used PATH Wave 7–Wave 4 cohort all-waves weights with Fay’s factor=0.3 for replicate weights.⁴⁴ To account for unequal spacing between PATH waves, we specified time intervals of two years for Waves 4–5 and 5–6 and one year for Waves 6–7 in the continuous-time Markov model. All analyses were performed in R (version 4.4.3) using the msm (version 1.8.2)²⁸ and wmsm (version 1.3)⁴⁵ packages. Detailed R code implementing the multistate modeling analyses (including model specification, estimation, and computation of transition probabilities) is available in the eAppendix.

RESULTS

Descriptive Analyses

Baseline characteristics of the study cohort (N=5,575; 50% female; 68.9% White; 21.6% Hispanic) are available in eTable 1. The none/low symptom state was the most prevalent at baseline, (62.1%, n=3,481), followed by the co-occurring symptoms state (15.0%, n=833), the exclusive internalizing state (12.1%, n=678), and the exclusive externalizing state (10.8%, n=583). Compared to male respondents, female respondents reported higher internalizing (14.5% vs. 9.8%) and co-occurring symptoms (16.5% vs. 13.5%) while males reported higher externalizing symptoms than females (12.4% vs. 9.2%). White participants reported higher co-occurring symptoms than non-White participants (16.5% vs. 11.6%). Substance use was highest among those participants with co-occurring symptoms and lowest among participants in the none/low state across all four substances (tobacco, alcohol, cannabis, prescription drugs).

Figure 2 shows the observed number of participants in each state from 2016 to 2023. Prevalences of internalizing (12.1%[95% CI=11.2–12.9] to 13.0%[95% CI=12.1–13.9]) and externalizing (10.4%[95% CI=9.6–11.2] to 8.7%[95% CI=8.0–9.5]) symptoms were relatively stable. We did not observe a clear temporal trend for none/low symptom prevalence (62.5%[95% CI=61.2–63.8] to 63.2%[95% CI=61.9–64.4]) or for co-occurring symptom prevalence (13.9%[95% CI=13.0–14.8] to 15.1%[95% CI=14.2–16.0]).

Figure 2.

Observed Transitions in Mental Health State Over Time, Population Assessment of Tobacco and Health Study, 2016–2023, N=5,575^a

^a Mental health states were mutually exclusive and assessed based on self-reported symptoms within the past 12 months. Internalizing and externalizing problems were coded as binary indicators: none/low (<4 symptoms in the past 12 months) versus high (≥4 symptoms in the past 12 months). The category ‘co-occurring mental health symptoms’ refers to the co-occurrence of internalizing and externalizing problems. The x-axis spacing reflects unequal time between PATH waves (2 years for Waves 4–5 and 5–6; 1 year for Waves 6–7). Consequently, empirical transition fractions are not directly comparable across intervals of different duration. Percentages (%) for each state from Wave 4 to Wave 7: none/low mental health symptoms (62.5, 64.1, 67.5, 63.2), exclusive internalizing symptoms (12.1, 12.6, 11.1, 13.0), exclusive externalizing symptoms (10.4, 9.4, 9.6, 8.7), and co-occurring mental health symptoms (15.0, 13.9, 11.9, 15.1).

Mental Health Transition Probabilities

Figure 3 depicts the estimated short-term and long-term transition probabilities across mental health states. Over two waves, most participants reporting none/low mental health symptoms at baseline remained in the same mental health state (79.9%[95% CI=78.8–80.9]). Among participants who were in the exclusive internalizing state at baseline, 29.1% (95% CI=26.8–31.5) remained in the same state at the subsequent assessment. Similarly, among participants who were in the exclusive externalizing state at baseline, 27.7% (95% CI=24.7–30.8) remained in that state at the next transition point. Of participants with co-occurrence of internalizing and externalizing symptoms at baseline, 43.9% (95% CI=40.1–46.8) remained in the co-occurrence state at the next assessment.

Figure 3.

Estimated Short-Term and Long-Term Transition Probabilities between Mental Health States, Population Assessment of Tobacco and Health Study, 2016–2023, N=5,575^a

^a Short-term and long-term cumulative transition probabilities refer to estimates of mental health transitions occurring one wave after baseline (at Wave 2) and three waves after baseline (at Wave 4). Mental health states were mutually exclusive and assessed based on self-reported symptoms within the past 12 months. Internalizing and externalizing problems were coded as binary indicators: none/low (<4 symptoms in the past 12 months) versus high (≥4 symptoms in the past 12 months). The category ‘co-occurring mental health symptoms’ refers to the co-occurrence of internalizing and externalizing problems. Values indicate cumulative transition probabilities (%), and row probabilities for each plot sum to 100. Estimates were weighted to generate nationally representative results, incorporating adjustments for the complex sampling design of the survey. Rows (Y-axis) represent baseline mental health states, and columns (X-axis) represent mental health states at follow-ups.

Transitions from none/low-symptom states were rare (<10%), and cross-domain shifts were uncommon (6.1%[95% CI=5.5–6.8] internalizing to externalizing and 7.7%[95% CI=7.1–8.4] in the reverse direction). While many with internalizing or externalizing symptoms transitioned to none/low symptoms ((43.2%, 95% CI=40.6–45.7 and 44.8%, 95% CI=41.2–48.4, respectively), nearly 20% each transitioned from exclusive internalizing or exclusive externalizing to the co-occurrence state. Observed transition frequencies and rates showed similar patterns (eTable 2).

Long-term estimates showed that approximately 60% of those in the exclusive internalizing, exclusive externalizing, or co-occurring symptom states at baseline transitioned to the none/low symptom state by Wave 7. Approximately 12–16% of participants in the none/low or exclusive internalizing or externalizing symptom states at baseline transitioned to the co-occurring symptom state by Wave 7. The estimated mean sojourn time for co-occurring symptom states was 0.92 (95% CI=0.83–1.01) years, compared to 0.64 years and 0.66 years for exclusive internalizing and externalizing states, respectively (eTable 3). The transition intensity matrix used to estimate short- and long-term transition probabilities is presented in eFigure 2.

eTable 4 reports ratios of transition intensities comparing analogous symptom transitions. Across comparisons, co-occurring symptoms were characterized by slower remission and substantially greater persistence compared to exclusive symptom states. Participants with exclusive symptoms more commonly transitioned into co-occurring symptoms than participants with none/low symptoms transitioned to an exclusive symptom state.

Substance Use and Mental Health Transitions

Figure 4 and eTable 5 show HRs for transitions between symptom states associated with substance use. Current established nicotine use was associated with a higher risk of transitioning from none/low symptoms to internalizing symptoms (HR=1.47, 95% CI=1.09–1.99). Past 30-day binge drinking (HR=1.76, 95% CI=1.30–2.40) and prescription drug misuse (HR=3.42, 95% CI=1.83–6.38) were associated with increased risk of transitioning from none/low symptoms to externalizing symptoms. Past 30-day cannabis use was associated with higher risk of transitions from none/low symptoms to both internalizing (HR=1.52, 95% CI=1.19–1.94) and externalizing (HR=1.63, 95% CI=1.24–2.14) symptoms. Substance use behaviors were not associated with transitions from exclusive symptom states to either none/low symptoms or co-occurring symptoms, nor with transitions from co-occurring symptoms to exclusive symptom states.

Figure 4.

Hazard Ratios of Transitions by Substance Use Behaviors, Population Assessment of Tobacco and Health Study, 2016–2023, N=5,575^a

^a Mental health states were mutually exclusive and assessed based on self-reported symptoms within the past 12 months. Internalizing and externalizing problems were coded as binary indicators: none/low (<4 symptoms in the past 12 months) versus high (≥4 symptoms in the past 12 months). The category ‘co-occurring mental health symptoms’ refers to the co-occurrence of internalizing and externalizing problems. Separate models were tested for each substance use behavior; each model simultaneously adjusted for sociodemographic covariates, including sex, race, and ethnicity. Due to insufficient case counts for certain transitions—transitions from none/low symptoms to co-occurring symptoms and from co-occurring symptoms to none/low symptoms, as well as transitions between internalizing and externalizing symptoms—these transitions were constrained to 0 in the transition matrix. Estimates were weighted to generate nationally representative results, incorporating adjustments for the complex sampling design of the survey. Nicotine = Current established cigarette smoking (current every-day or some-day use of cigarettes and had smoked more than 100 cigarettes in their lifetime) and/or e-cigarette use (current every-day or some-day use of e-cigarettes and had used electronic nicotine products fairly regularly); Binge drinking = Past 30-day binge drinking (four or more drinks for female and five or more drinks for male respondents on the same occasion); Cannabis = Past 30-day cannabis use, including hash, tetrahydrocannabinol (THC), grass, pot, weed or cigar/cigarillo/filtered cigar containing marijuana; Prescription drug = past 30-day prescription drug (stimulants such as Adderall and Ritalin, painkillers, sedatives or tranquilizers) misuse.

Sensitivity and Supplementary Analyses

Sensitivity analyses using unweighted estimates yielded results largely consistent with the primary analyses (eFigure 3). The estimated short-term transition probabilities were largely consistent with the corresponding empirical transition proportions (eFigure 4). As shown in eTable 6, e-cigarette use was associated with transition from none/low to internalizing symptoms (HR=1.59, 95% CI:1.05–2.40) while cigarette smoking was not. In contrast, cigarette smoking was associated with transition from externalizing to co-occurring symptoms (HR=2.14, 95% CI:1.09–4.20) while e-cigarette use was not. Supplementary analyses examining HRs of transitions by sociodemographic characteristics (eTable 7) suggested that sex (female vs. males) was positively associated with transition from the no/low-symptom state to the internalizing state (HR=1.27, 95% CI=1.03–1.57) and from the externalizing state to the co-occurring state (HR=1.46, 95% CI=1.23–1.89).

DISCUSSION

In this prospective cohort study (2016–2023), most young adults were in the none/low symptom state at baseline and remained in this state through subsequent waves. In short-term analysis, slightly less than half of participants in exclusive symptom states (internalizing or externalizing) at baseline transitioned to the none/low symptom state by the end of the study period, but roughly 20% of participants in each of these groups (exclusive externalizing or internalizing symptom states) transitioned to the co-occurring symptom state over the same time frame. A substantial proportion (≥40%) of participants in the co-occurring symptom state at baseline remained in this state by the end of the study period. Fewer than 10% of participants in the none-low symptom state at baseline transitioned to any symptomatic state (exclusive or co-occurring) over the study period. In long-term estimations, over 10% of those with internalizing or externalizing symptoms persisted in the same state, ~15% transitioned to co-occurring symptoms, and over 20% of those with none/low symptoms transitioned into any symptomatic state. Substance use was associated with transitions to adverse mental health states.

Transitioning to a co-occurring symptom state (approximately 20% and 15% of participants in exclusive symptom states at baseline in short- and long-term estimates, respectively) may be clinically meaningful because this co-occurring state may persist longer than exclusive symptom states.⁴⁶ Co-occurring symptom states are also associated with more adverse outcomes than exclusive symptom states.²¹ Direct transitions between exclusive internalizing and externalizing symptom states were relatively uncommon in our study, which is congruent with prior work.²⁰ This may reflect distinct etiological pathways for internalizing and externalizing symptom states, with internalizing symptoms linked to stress sensitivity and negative affect and externalizing symptoms to disinhibition and reward sensitivity.⁹ Our results suggest that transition from an exclusive symptom state to a co-occurring symptom state may be more common than transition from one exclusive symptom state to another exclusive symptom state.

We observed associations between substance use and transition probabilities from the none/low symptom state to one of the exclusive symptom states. In our results, e-cigarette use—but not cigarette smoking—was positively associated the transition from the none/low symptom state to the exclusive internalizing symptom state, which is somewhat incongruent with prior evidence and existing theoretical frameworks.⁴⁷ While some evidence indicates that nicotine affects the brain’s serotonin and dopamine systems—central to mood regulation and stress sensitivity,^48,49 there is little evidence for an independent effect of nicotine on depression risk.⁵⁰ One possible explanation for our finding is that this cohort of young adults came of age during a period in which e-cigarettes, rather than combustible cigarettes, were the predominant nicotine product. As such, e-cigarette use might better capture contemporary patterns of nicotine use in terms of initiation, frequency, dependence trajectories, and underlying psychosocial vulnerability. Interestingly, cigarette smoking, but not e-cigarette use, was associated with the transition from externalizing symptoms to co-occurring symptoms, which could suggest that rather than indicating a direct biological effect, cigarette smoking serves as a marker of broader behavioral risk and vulnerability. Binge drinking and prescription drug misuse were associated with transitions to externalizing symptoms, consistent with the literature documenting these substance use behaviors in association with behavioral disinhibition, heightened impulsivity, and the social reinforcement of risk-taking behaviors.⁵¹ Cannabis use was associated with transitions to both internalizing and externalizing symptoms, which may reflect the multiple ways cannabinoids influence stress, mood, and cognitive control⁵² as well as variations in use contexts (e.g., social vs. solitary).^53,54

Substance use was not associated with transitions from exclusive or co-occurring symptom states to the none-low symptom state despite the common perception among young adults that substances can be used as a coping strategy,⁵⁵ which is consistent with prior evidence.^56,57 Our results indicate that substance use may be associated with transitions from the none/low symptom state to internalizing or externalizing symptom states, suggesting that substance use may be associated with the onset or re-emergence of mental health symptoms. However, we did not find substance use behaviors to be associated with transitions from exclusive symptom states to co-occurring symptom states, which might possibly support the hypothesis that substance use behaviors are more likely to exacerbate existing mental health symptoms than to promote the emergence of new symptoms. However, this interpretation is highly speculative and was not directly evaluated in the present study. This question warrants further investigation in future research.

We also found sex to be associated with transitions between mental health symptom states. Compared to males, female participants were less likely to transition from the exclusive internalizing symptom state to the none/low-symptom state more likely to transition from the exclusive externalizing symptom state to the co-occurring symptom state. These results may be related to both biological (endocrine) and psychosocial factors. In general, females exhibit greater stress sensitivity than males; hormonal regulation of serotonergic and dopaminergic pathways may be related to the persistence of internalizing symptoms.^58,59 Psychological research also suggests that females are more likely than males to rely on ruminative coping styles,⁶⁰ which could partially explain for the persistence of internalizing symptom states and the prevalence of co-occurring externalizing and internalizing symptoms states.⁶¹ Young adult females may face unique risks for both the persistence of internalizing problems and the development of co-occurring symptom profiles.

This study had several limitations. Internalizing and externalizing symptom measures were dichotomous, limiting our ability to assess symptom frequency and intensity. The Markov approach assumes transitions depend only on the current state,²⁸ although certain risk factors (e.g., previous prolonged substance use) may predict future mental health problems, even in the absence of recent use. This study used observed mental health states based on self-reported symptom measures. Although hidden Markov models can account for potential misclassification by modeling the unobserved true states through error-prone indicators,⁶² we used a standard Markov modeling approach given that state membership was treated as known rather than latent. Relatedly, substance use and mental health symptoms were both assessed through self-reported measures at each survey wave, limiting our ability to determine temporal precedence. In this study, residual confounding cannot be ruled out, as unmeasured factors (e.g., biological features) may influence both substance use and mental health transitions despite covariate adjustment. Residual confounding may bias estimates in either direction, potentially leading to over- or underestimation of effect measures. Attrition across survey waves may also introduce selection bias if individuals lost to follow-up differed from those retained, potentially affecting transition estimates despite weighting adjustments. If higher-risk participants were more likely to drop out, estimates may be attenuated; differential retention could inflate estimates. Although this study focused on how substance use may be associated with mental health transitions across waves, the reverse direction is also plausible, circumscribing causal interpretation of our results. Additional limitations include reliance on annual assessments, potential selection bias from attrition despite weighting adjustments for nonresponse, the inability to distinguish between cigarettes and e-cigarettes due to the small cell size of cigarette use in this young adult sample, and the inability to explicitly examine recurrence of mental health symptoms across multiple cycles (e.g., low–high–low patterns) within the constraints of the multistate Markov modeling framework.

CONCLUSIONS

In this nationally representative cohort of young adults in the U.S., substantial proportions transitioned from none/low-symptom state to co-occurring internalizing and externalizing symptoms in both short- and long-term analyses, and a substantial proportion of young adults with co-occurring symptoms at baseline remained in this symptom state over time. This study further suggested associations between substance use the probabilities of transitioning from a none/low-symptom state to internalizing or externalizing symptom states.

Data and Computing Code Availability:

Data are available in a public, open access repository. Data of the Population Assessment of Tobacco and Health are publicly available at https://www.icpsr.umich.edu/web/NAHDAP/studies/36498. The code used in this study is provided in the eAppendix.