Unhealthy behaviors associated with mental health disorders: a systematic comparative review of diet quality, sedentary behavior, and cannabis and tobacco use

Abstract

Background:

There are well-established literatures documenting the associations between mental disorders and unhealthy behaviors such as poor diet quality, sedentary behavior, and cannabis and tobacco use. Few studies have attempted to understand the respective findings in light of each other, however.

Objective:

The purpose of this review was to assemble comparable data for each behavior-disorder association and assess the associations in terms of their overall strength. The review aimed to include a representative, but not exhaustive, range of studies that would allow for explorative comparisons.

Methods:

Eligible studies were identified via Pubmed searches and citation searching, restricted to publications no older than 2015 written in English. To obtain comparable data, only studies that reported findings as odds ratios were included, and risk of bias related to study samples, behavioral measurement disparities, and control variables was assessed via sensitivity analyses. Findings for each disorder were compared on the basis of different measures of central tendency.

Results:

From 3,682 records, 294 studies were included. The review found evidence of associations between each of the four unhealthy behaviors and psychosis, depression, anxiety, bipolar disorder, attention-deficit/hyperactivity disorder (ADHD), and post-traumatic stress disorder (PTSD), while personality disorder was only investigated in relation to cannabis and tobacco use. In overall comparison, the associations were generally of similar strength, and only the association between cannabis use and personality disorder was exceptional in terms of being significantly stronger than its counterparts across disorders and across behaviors. Analyses of bias risk identified some influence from behavioral measurement disparities and lack of adequate statistical control, but findings were generally robust across a range of sensitivity analyses.

Conclusion:

This explorative and comparative review found that poor diet quality, sedentary behavior, and cannabis and tobacco use are about equally strongly associated with a range of different mental disorders. Given the general nature of these associations, we should probably understand them to reflect a general and shared etiology. However, the findings in this review should be regarded as tentative until confirmed by more comprehensive investigations.

Highlights

• In terms of mental disorder, moderate cannabis use appears to be about as harmful as moderate tobacco use.

• In terms of mental disorder, moderate cannabis use appears to be about as harmful as watching TV for more than 2–4 h per day or eating fast food and drinking sugary beverages every day.

• The association between cannabis use and psychosis is not stronger than the associations between cannabis use and other mental disorders.

• The association between psychosis and cannabis use is not stronger than the associations between psychosis and poor diet quality, sedentary behavior, or tobacco use.

Introduction

Associations between unhealthy behaviors and mental disorders are well established in research literature. This review considers four different unhealthy behaviors—cannabis use, tobacco use, sedentary behavior, and poor diet quality—and a range of mental disorders across the range from common mental disorders (CMDs) to serious mental illnesses (SMIs) and including both internalizing and externalizing disorders. Aiming for a broad but not exhaustive range of disorders that can provide a convincing basis for comparison, the seven disorders psychosis, depression, anxiety, bipolar disorder, personality disorder, attention deficit/hyperactivity disorder (ADHD), and post-traumatic stress disorder (PTSD) were included in the review.

Psychosis

Several recent reviews have confirmed an association between psychosis and cannabis use (1–6), while a few have failed to identify a statistically significant relationship (7, 8). The strength of associations identified in meta-analyses ranged from strong for the most severe cannabis users [(5): OR 3.90] to moderate [(8): OR 1.786; (3): RR 1.71; (4): OR 1.75 for abuse/dependence; (5): OR 1.97 for any use] or weak [(7): RR 1.11; (4): OR 1.14 for lifetime use]. Several recent reviews have also confirmed an association between psychosis and tobacco use (8–11), with the strength of the association ranging from moderate to strong in meta-analyses (respectively, OR 2.22, 3.04, 3.22/2.18 for case-control/prospective studies, and RR 1.99). One review did not find evidence of any significant association (12). Those reviews that have compared the strength of the relationship between cannabis and psychosis to that between tobacco and psychosis have found that the latter appears to be somewhat stronger (8, 13). The associations between psychosis and poor diet quality or sedentary behavior have not been extensively studied, although reviews have suggested that people with psychosis tend to have poor diet quality (14, 15) and low levels of physical activity (16, 17).

Depression

A number of recent review articles support an association between cannabis use and depression, although with some mixed results (12, 18–28), while one found that the relationship is unclear (29). The meta-analysis by Onaemo et al. (26) identified a strong association between cannabis use disorder and major depression (OR 3.22), while those by Esmaeelzadeh et al. (18) and Gobbi et al. (19) as well as an earlier meta-analysis by Lev-Ran et al. (30) identified a weak to moderate association (respectively, OR 1.29, 1.37 and 1.62) for overall cannabis use.

For tobacco, recent reviews by Weinberger et al. (31) and Fluharty et al. (32) confirmed that people with depression are more likely to smoke and to meet criteria for nicotine dependence, with the latter also indicating that smoking may lead to depression. A meta-analysis by Esmaeelzadeh et al. (18) found that the association between tobacco use and depression was moderately strong with an OR of 1.65, which was stronger than that for the corresponding association between cannabis use and depression. Another meta-analysis by Groenman et al. (33) found that childhood depression was a risk factor for subsequent nicotine use disorder (OR 2.56). Reviews by Zeng and Li (34) and Han et al. (35) also confirmed that secondhand smoking is associated with depression, with their meta-analyses obtaining ORs of 1.60 and 1.32, respectively. Furthermore, recent reviews have also indicated that e-cigarette use is associated with depression (36, 37).

There are also a number of reviews supporting associations between depression and sedentary behavior (17, 38–47) and poor diet quality (48–57), although with some mixed results. In meta-analyses, the associations were weak to moderate in strength, with RR 1.42 (47) and RR 1.25 (46) for overall sedentary behavior, OR 1.28 (45) and OR 1.12 (42) for screen time, RR 1.18 for TV viewing (41), OR 1.44 for ultra-processed food consumption (49), RR 1.25 (58) and RR 1.31 (48) for sugar-sweetened beverage consumption, OR 1.18 for western-style unhealthy dietary pattern (59), and OR 1.62 (52) and OR 1.11 (54) for fast/junk food.

Anxiety

Some reviews have identified an association between cannabis use and anxiety (18, 20, 26, 27, 60–62), although others have maintained that the relationship is unclear (19, 22, 28, 29, 63, 64). In meta-analyses, moderate to strong associations were identified between cannabis use disorder and anxiety [(60): OR 1.68; (26): OR 2.99], while the associations for overall use were generally weak [(18): OR 1.36; (19): OR 1.18; (60): OR 1.24; (61): OR 1.15; (62): OR 1.25]. Reviews have also identified an association between tobacco use and anxiety (12, 18, 65–67), although with some mixed results. A meta-analysis by Esmaeelzadeh et al. (18) found a strong association between tobacco use and anxiety with an OR of 2.21, which was stronger than that for the corresponding association between cannabis use and anxiety.

Reviews have also found evidence of an association between anxiety and sedentary behavior (68–71) and poor diet quality (49, 52). Associations in meta-analyses were of weak to moderate strength, with regular physical activity reducing the odds for anxiety [(72): OR 0.54; (70): OR 0.74] and sedentary behavior [(68): OR 1.48], ultra-processed food [(49): OR 1.48] and junk food [(52): OR 1.24] increasing the odds for anxiety.

Bipolar disorder

The relationship between cannabis use and bipolar disorder has not been intensively studied, but a number of reviews have suggested a positive association (20, 23, 25, 73–78). Meta-analyses by Gibbs et al. [(73): OR 2.97] and Hunt et al. [(74): OR 2.35] identified a strong association between cannabis use disorder and bipolar disorder. Some reviews have also suggested an association between tobacco use and bipolar disorder (75, 79), although the evidence is mixed (12). Nevertheless, the meta-analysis by Jackson et al. (79) identified a strong association with an OR of 3.5. Reviews have also identified tentative associations between bipolar disorder and sedentary behavior (17, 80, 81) and diet quality (82, 83), but there are few available studies.

Personality disorders

While the association between personality disorders and general substance use is well established, the evidence specifically for cannabis (20, 28, 84–86) and tobacco (12, 87) is quite limited. Similarly, the research into the relationships between personality disorders and sedentary behavior and diet quality is very limited (87, 88).

Attention-deficit/hyperactivity disorder

ADHD is commonly diagnosed in childhood, and most studies investigating the relationship between the disorder and substance use focus on ADHD as a risk factor for subsequent substance use. For cannabis, meta-analyses have found strong associations between childhood ADHD and regular [(89): OR 2.45] and lifetime use [(90): OR 2.78] as well as a moderate association for abuse or dependence [(90): OR 1.58]. For tobacco, meta-analyses similarly identified childhood ADHD as a strong risk factor both for subsequent nicotine use [(89) OR 2.16; (90): OR 2.08] and nicotine dependence or use disorder [(33): OR 2.52; (90): OR 2.82]. The evidence for e-cigarette use is very limited (36).

Reviews have also supported associations between ADHD and sedentary behavior (91–95) and diet quality (96–99). Meta-analyses of sedentary behavior found moderate to strong associations between electronic media use and ADHD [(91): OR 1.94; (92): 2.597], while meta-analyses of diet quality found weak to strong associations for sugar and soft drink consumption [(97): pooled effect size 1.22], junk food [(99): OR 1.51], and general unhealthy diets [(96): OR 1.41; (98): OR 2.24].

Post-traumatic stress disorder

Reviews have identified an association between cannabis use and PTSD (100, 101) and investigated the potential for therapeutic use (102–104). Reviews have also identified an association between tobacco use and PTSD (105–109), with one meta-analysis by van den Berk-Clark et al. (109) finding a strong association (OR 2.13) for unadjusted studies and a weak association (OR 1.22) for a single adjusted study. The evidence for e-cigarette use is very limited (36).

Furthermore, reviews have suggested that both sedentary behavior and poor diet quality are associated with PTSD, although with some mixed results (70, 109–111). In meta-analyses, high physical activity was moderately protective against PTSD [(70): OR 0.57], while PTSD in turn was a weak risk factor for low physical activity [(109): OR 0.91 for adjusted studies and OR 0.69 for unadjusted studies]. The meta-analysis by van den Berk-Clark et al. (109) on diet quality was somewhat unclear, as four unadjusted studies predicted a healthier diet (OR 1.25) while a single adjusted study predicted a less healthy diet (OR 0.95).

While these reviews indicate a broad range of associations between unhealthy behaviors and mental disorders, they also demonstrate that such associations have not often been understood in relation to each other across various behaviors and disorders, which would allow for both a calibration of their relative importance and a widened perspective on their etiology. Instead, individual associations are usually analyzed and discussed in isolation, which encourages the perspective that the association under scrutiny is unique and reflects a particular and individual etiology. Should it be true that the range of disorders here under scrutiny are generally associated with a range of unhealthy behaviors related to diet, exercise, and licit and illicit substance use, however, it might seem unlikely that these parallel associations are best understood in isolation from one another. If the association between a given behavior and a given mental disorder has parallels across a range of disorders and behaviors, its lack of specificity indicates that the association probably does not reflect a causal effect from the behavior itself (112). Instead, such a field of associations most likely point to general and shared dynamics. One such dynamic may involve underlying issues of social misery and psychological trauma that perhaps give rise to both unhealthy behaviors and to mental disorders. To take one example, long-term unemployment may entail poverty, boredom, and social isolation, which in turn may push a person into unhealthy behaviors as means of escapism and diversion and also contribute to a mounting sense of hopelessness and depression.

Indeed, there is strong evidence supporting a relationship between mental disorder and factors relating to misery and trauma. To take a few noteworthy examples, childhood abuse has been associated with psychosis at ORs of 7.3 (113) and 5.9 (8), with depression at ORs of 19.1 (114) and 4.2 (115), and with ADHD at ORs of 7.3 (91) and 3.3 (116), as well as with cannabis abuse at ORs of 5.8 (117), 2.7 (118), and 2.2 (119). Violence victimization has similarly been associated with psychosis at ORs of 5.9 (120), 4.0 (121), 2.2 (122), and 1.9 (123), with PTSD and depression at ORs of 32.4 and 5.4 (124), and with nicotine dependence at ORs of 22.4 (124) and 5.5 (125). Afifi et al. (118) for their part found that living in an unsafe community was associated with cannabis use at an OR of 5.0 and with tobacco use at an OR of 3.8. Comparable findings exist for childhood bullying victimization (120, 126), while poverty, low education, and unemployment have been associated with psychosis at ORs of 8.2 (127) and 3.0 (122), with depression at ORs of 9.1 (128), 5.0 (129), 3.4 (130), 3.2 (131), 2.8 (132), and 2.6 (133), with nicotine dependence at ORs of 37.0 (134), 2.4 (135), and 2.3 (125), and with cannabis use disorder at an OR of 2.5 (136). It should also be noted that factors relating to trauma and misery might accumulate in an individual and thereby increase risk. In a study by McMahon et al. (120), cumulative adverse life events were strongly associated with psychotic experiences, with four or more adverse life events incurring an OR of 16.8.

Thus, it seems possible that both unhealthy behaviors and mental disorders might be caused by underlying problems related to social misery and psychological trauma, and that the observed associations between those behaviors and disorders are largely spurious. As a basis for the discussion of the specificity of individual associations, this article will use the probably most discussed associations among those here under purview, namely that between cannabis use and psychosis, which is commonly taken to reflect a causal effect from cannabis use. A recent consensus paper by D'Souza et al. (1) pointed to a study by Starzer et al. (137) as evidence supporting the specificity of this association, and on this basis drew the inference that “[c]annabis is more likely to be associated with psychosis outcomes than other psychiatric diagnoses” (p. 732). This consensus paper also pointed to a somewhat broader range of evidence to support the specificity of the cannabis-psychosis association as compared to other forms of substance use, drawing the inference that “[t]he risk for a psychosis outcome is highest for cannabis relative to other drugs” (ibid.).

However, there is also evidence contradicting the consensus view presented in D'Souza et al. (1). As Ksir and Hart (138) pointed out, research has associated cannabis use with depression, which may challenge the specificity of the cannabis-psychosis association. Furthermore, a recent review by Johnstad (13) found that tobacco use tended to be at least as strongly associated with psychosis as is cannabis use. The cannabis-psychosis association may therefore lack specificity both across disorders and across behaviors, which would open for the perspective that the association may be spurious and reflective of a general etiology perhaps related to underlying psychological trauma and social misery.

This interpretation is supported by a study by Shakoor et al. (139) which found that in a sample of adolescent twins, “[e]nvironmental influences explained all of the covariation between cannabis use and paranoia, cognitive disorganization and parent-rated negative symptoms,” with a bivariate common environment of 69%−100% (p. 144). Cannabis use explained 2%−5% of variance in positive, cognitive, and negative psychotic experiences. Sideli et al. (140) for their part found that less than daily cannabis use was not associated with psychosis (OR 0.97, p = 0.92) unless it co-occurred with having experienced child abuse, in which case the association was strong (OR 2.51, p < 0.05), while Copeland et al. (141) found that cannabis use before the age of 16 was strongly associated with depression (OR 2.3) and anxiety (OR 2.9) in an analysis adjusted only for sex and race/ethnicity, but that the association disappeared with adjustment for childhood disorders and adversities (OR 0.6 for both depression and anxiety). A meta-analysis by Fusar-Poli et al. (8) similarly found that risk for psychosis was related to environmental risk factors such as childhood trauma and further accumulated by low education and unemployment, while finding no significant impact from cannabis use.

In sum, the worry that unhealthy behaviors such as cannabis use cause psychosis may be a case of projecting the social characteristics of the drug-using population onto the drug itself and understand these characteristics as pharmacological effects inherent to the drug use. This comparative review aimed to assemble data on a broad range of associations between unhealthy behaviors and mental disorders in order to compare their respective strengths and to gain insight into the influence of factors relating to trauma and misery by recording the inclusion of relevant control variables and subsequently running sensitivity tests on the dataset. It reviews recent (2015 through January 2023) research on associations between a range of unhealthy behaviors and mental disorders, using the much-debated association between cannabis use and psychosis as a basis for comparison. In order to obtain comparable figures, the review included only records that presented their findings as odds ratios, and attempted to account for variations in behavior intensity, sample generalizability, and covariates via subsequent sensitivity analyses. By thus gaining a comparative overview of the relationship between unhealthy behaviors and mental disorders, the review hoped to achieve three specific aims:

• Understand individual associations between a given unhealthy behavior and a given mental disorder in relation to other such disorders.

• Understand individual associations between a given unhealthy behavior and a given mental disorder in relation to other such behaviors.

• Understand the association between unhealthy behaviors and mental disorders in relation to statistical control for covariates.

Methods

The pre-planned criteria for inclusion were that studies must investigate associations between any of the relevant behaviors and disorders and report their findings as odds ratios in a manner that allows for cross-study comparisons. In order to keep the number of included studies manageable, only recent studies (no older than 2015) would be included. Furthermore, the studies should be published in English, be available in the Pubmed database, and their sample size should be at least 100.

Two Pubmed searches were performed (February 3, 2023). The search related to cannabis and tobacco use was [(tobacco OR nicotine OR cigarette OR cannabis OR marijuana) AND (psychosis OR depression OR anxiety OR bipolar OR “personality disorder” OR adhd OR ptsd) AND associat^* AND (use OR “use disorder” OR dependence OR abuse) AND (comorbid OR dual OR psychiatric)] while the search related to diet quality and sedentary behavior was [(“poor diet” OR “sedentary behavior” OR “fast food” OR soda OR “sugar-sweetened”) AND (psychosis OR depression OR anxiety OR bipolar OR “personality disorder” OR adhd OR ptsd) AND associat^*]. The author assessed all records. Reported associations in included studies were classified as cross-sectional or longitudinal, with retrospective analyses being classified as longitudinal. In order to investigate directionality as openly as possible, longitudinal studies of both unhealthy behaviors leading to mental disorders and mental disorders leading to unhealthy behaviors were included. Study quality was assessed on the level of sample representativeness, behavioral measurement disparities, and the inclusion of control variables related to social misery and psychological trauma. Results are reported according to PRISMA guidelines (142).

The selection procedure is shown in the flow chart of Figure 1. Criteria related to language and publication date were implemented in the search itself. Four reports were excluded for having samples lower than n = 100, while 309 reports were excluded because they did not report any relevant odds ratios that allowed for comparisons. This was related to the following reasons:

• The reports combined several types of drugs or disorders into a single construct (e.g., “substance use,” “psychological distress,” “internalizing disorders”).

• The reports subdivided disorders into specific symptoms or symptom classes (e.g., auditory and visual hallucinations in psychosis).

• Findings were reported in other terms than odds ratios (e.g., risk ratios, hazard ratios).

• Some reports especially of sedentary behavior were also excluded because they reported ORs for continuous variables that were not readily comparable to the other studies included in the review.

• A few reports utilizing Mendelian randomization methodology were excluded because they were not comparable to the cross-sectional or longitudinal studies included in the review and did not constitute a meaningful basis for comparison on their own.

Figure 1

Variables

The review focused on moderate cannabis and tobacco use as compared to high levels of sedentary behavior and poor diet quality. This approach was founded on the observation in Johnstad (13) that heavy substance use is very likely associated with underlying factors of social misery and psychological trauma, and that analyses of the heavy drug-using population are therefore at high risk of confusing the social characteristics of this population segment with the effects from the drug use itself. While heavy chronic drug use is commonly driven by coping motives related to underlying misery and trauma, moderate use is driven by a much broader range of motivations (143–145). In order to understand pharmacological effects from substance use in and of themselves, it therefore seems advisable to focus on moderate usage patterns.

In this review, moderate cannabis use was defined as one or two use occasions per week (146, 147). For tobacco, moderate use could be defined as daily use of about 10 cigarettes (148–150). When the studies included in this review reported effects related to several different usage patterns, the one most closely resembling a moderate pattern was selected. The review did not exclude any study on the basis of divergences in the usage pattern, however, and several included studies focused exclusively on use disorders or similarly problematic behavior. For sedentary behavior and diet quality, the review focused on high exposure represented by such measures as the worst quartile, daily consumption of soda or fast-food, not meeting physical activity guidelines, or more than 4–8 h of screen time per day. This approach was based on the observation that unlike cannabis and tobacco use, sedentary behavior and unhealthy food consumption are universal behaviors with which moderate levels of engagement are ubiquitous. When studies reported figures from several different multivariate models, the review included the figures from the final, most stringently adjusted model. Sensitivity analyses attempted to account for the inclusion of studies focusing on specialty samples and disordered behavior patterns.

Analyses

Across studies

The meta-analysis compared mean odds ratios for associations between behaviors and disorders across studies. To compare odds ratio means related to a given unhealthy behavior and a given mental disorder in relation to other such disorders, the data was restructured so as to combine odds ratios for different disorders into a single variable with different corresponding labels. T-tests were used for each pair of disorders for cross-sectional and longitudinal (behavior before disorder) studies. With 21 pairs of disorders, 4 behaviors, and 2 analytical approaches, a total of 168 different comparative analyses were performed. To compare odds ratio means between different unhealthy behaviors, findings for each disorder were compared for each pair of behaviors for cross-sectional and longitudinal (behavior before disorder) studies. With seven disorders, six pairs of behaviors, and two analytical approaches, a total of 84 different comparative analyses were performed between studies of different behaviors. It might be noted that none of these comparisons were Bonferroni corrected, which would probably have rendered them all non-significant.

Within studies

Some studies reported comparable odds ratios for several different disorders and/or behaviors. In studies of cannabis use that reported figures relating both to psychosis and to other disorders, the review compared the respective odds ratios and classified the figure for psychosis as stronger if it exceeded the other figure by 0.25, as weaker if the other figure exceeded the figure for psychosis by 0.25, and otherwise as being at the same level. Similarly, for studies that provided odds ratios relating to a given disorder for both cannabis/tobacco use and diet quality/sedentary behavior, the review classified either as stronger if it exceeded its counterpart by 0.25, and otherwise as being at the same level.

Sensitivity analyses

Sample groups

The studies included in the review analyzed participant samples ranging from representative samples of the nation's adult population to more narrowly defined groups such as children, older adults, male military conscripts, university students, and patient groups. In order to control for the possibility that narrow samples unduly impacted overall dataset analyses, dichotomous variables for “children” (samples below age 18), “older adults” (samples above age 60), “patient” (all patient or high-risk groups without controls), and “special” (samples consisting of university students or employees, all-male or all-female samples, military service members or veterans, and civil servants) were defined. Subsequently, these variables were combined into a single dichotomous indicator for specialty sample.

Behavioral measurement disparities

The studies included in the review spanned behaviors across the risk spectrum, from any (or ever) use of tobacco or weekly fast-food consumption to nicotine use disorder and binge eating disorder. In order to control for the possibility that the disordered behavior measured in some studies unduly impacted overall analyses, a dichotomous “problem behavior” indicator was defined to include behavioral variables including “binge eating disorder,” “emotional eating,” “maladaptive/pathological internet use,” “smartphone addiction,” “Facebook overuse,” “problematic gaming,” “internet addiction,” tobacco/cannabis problem use, abuse, dependence, use disorder, or high-risk use, and smoking during pregnancy. Especially for tobacco, however, it was not clear whether use measured in terms of daily smoking could be meaningfully differentiated from use measured in terms of abuse, dependence etc.

Control variables

The studies included in the review used a wide variety of control variables in their statistical analyses. In this review, systematic differences in association strengths related to statistical control for diet quality or sedentary behavior (for tobacco and cannabis analyses), tobacco use (for diet quality, sedentary behavior, or cannabis analyses), and cannabis use (for diet quality, sedentary behavior, or tobacco analyses) were controlled for by the application of dichotomous adjustment variables (0 = no control, 1 = control). Furthermore, dichotomous variables for adjustment for social misery (1 = inclusion of any control variable relating to poverty, education, or unemployment) and psychological trauma (1 = inclusion of any control variable relating to childhood abuse/neglect, bullying victimization, or violence victimization) were defined. Finally, a variable for studies that reported bivariate associations (1 = bivariate) was defined.

It should be noted, however, that some studies analyzed potential covariates on a bivariate basis and only included those that passed a significance threshold of p < 0.05 in the final model. This review accepted such variables as having been controlled for, although it should be clear that a covariate may exert meaningful influence on the outcome of a multivariate regression even if its bivariate correlation with the outcome variable is significant only at the 90% (p < 0.1) level, for instance.

Results

Overall findings

From 3,682 records, 294 were included. Some publications reported figures for several behaviors and methodological approaches (commonly, both cross-sectional and longitudinal), and counting these the review included 97 studies of cannabis use, 132 studies of tobacco use, 88 studies of sedentary behavior, and 47 studies of diet quality. The dataset included a total of 570 associations between behaviors (cannabis use, tobacco use, sedentary behavior, and diet quality) and disorders (psychosis, depression, anxiety, bipolar disorder, personality disorder, ADHD, and PTSD).

Sample groups

The review included 320 associations (out of 570) marked as relating to specialty samples. In the full dataset of cross-sectional and longitudinal associations, the specialty sample indicator was not significantly correlated with odds ratios for any disorder or behavior. Specialty samples were most commonly used in studies of diet quality and sedentary behavior (69%) and least commonly used in cannabis studies (44%). As a sensitivity test, the analyses reported below were repeated with studies using specialty samples removed from the dataset. This new set of analyses identified marginally significant differences (t = 2.30, df = 5.76, two-sided p = 0.06) for PTSD between cannabis use and tobacco use, with the former having higher ORs.

Behavioral measurement disparities

The review included 101 associations (out of 570) marked as relating to disordered or problematic behavior. In the full dataset of cross-sectional and longitudinal associations, the problem behavior indicator was significantly associated with odds ratios for depression (r = 0.15, p = 0.015, N = 251) and ADHD (r = 0.31, p = 0.021, N = 55). The inclusion of studies investigating problem behaviors may thus have served to increase association strengths for these disorders. Across disorders, the indicator for problem behavior was significantly correlated with the odds ratio strength in non-tobacco associations (r = 0.12, p = 0.021, N = 366) but not in associations with tobacco use (r = 0.08, p = 0.252, N = 204). This finding supports the notion that tobacco use measured in terms of daily smoking cannot necessarily be differentiated from use measured in terms of abuse or dependence. Problem behaviors were most commonly studied in relation to cannabis use (29%) and least commonly in relation to diet quality and sedentary behavior (10%). As a sensitivity test, the analyses reported below were repeated with studies investigating disordered or problematic behavior removed from the dataset. This new set of analyses identified significant differences (t = 2.31, df = 8.38, two-sided p < 0.05) for PTSD between cannabis use and tobacco use, as well as between cannabis use and sedentary behavior (t = 2.32, df = 8.13, two-sided p < 0.05), with ORs for cannabis in both cases being higher. Furthermore, these analyses identified marginally significant differences (t = 1.93, df = 32, two-sided p = 0.06) between cannabis-psychosis and cannabis-PTSD, with the latter having higher ORs. Finally, the review included a few studies that investigated the association between depression and secondhand smoking (N = 5) and e-cigarette use (N = 6). These studies did not diverge substantially from other tobacco studies in terms of their association strengths with mental disorders.

Control variables

Some studies reported bivariate associations (33 out of 570), generally because their main focus lay elsewhere and the relationships investigated in this review were only a secondary concern in their analysis. In the full dataset, the bivariate indicator was significantly associated with odds ratios for anxiety (r = 0.19, p = 0.043, N = 112) and ADHD (r = 0.33, p = 0.013, N = 55). The inclusion of studies reporting bivariate associations may thus have served to increase association strengths for these disorders. Across disorders and behaviors, the bivariate indicator was marginally correlated with odds ratios (r = 0.08, p = 0.064, N = 570). Bivariate findings were most commonly reported in tobacco studies (8%) and least commonly in studies of sedentary behavior and diet quality (4%). As a sensitivity test, the analyses reported below were repeated with studies reporting bivariate associations removed from the dataset. This new set of analyses identified marginally significant differences (t = 1.96, df = 18.57, two-sided p = 0.07) for PTSD between cannabis use and sedentary behavior, with the former having higher ORs.

Overall, 51 (out of 570) associations were adjusted for psychological trauma, and the indicator for trauma adjustment was marginally significant for psychosis (r = −0.25, p = 0.056, N = 61). Across disorders and behaviors, the indicator was marginally correlated with odds ratios (r = −0.07, p = 0.103, N = 570), and such adjustment was most commonly employed in cannabis studies (14%) and least commonly in studies of diet quality (2%). By contrast, 441 (out of 570) associations were adjusted for social misery, and such adjustment was most commonly employed in studies of diet quality (82%) and least in studies of cannabis and tobacco (76%). Across disorders and behaviors, the indicator for misery adjustment trended weakly in a negative direction (r = −0.06, p = 0.151, N = 570) but was marginally correlated with odds ratios in cannabis studies (r = −0.13, p = 0.093, N = 178).

Only 1% of cannabis studies and 14% of tobacco studies controlled for diet quality and/or sedentary behavior, while studies of the latter behaviors often controlled for tobacco (55%) but rarely for cannabis (8%). Furthermore, 35% of cannabis studies controlled for tobacco while 24% of tobacco studies controlled for cannabis. With regard to specific disorders, the indicator for cannabis adjustment was significant only in (non-cannabis) studies of psychosis (r = −0.55, p < 0.001, N = 32) while tobacco adjustment was significant in (non-tobacco) studies of anxiety (r = −0.24, p = 0.033, N = 80) and marginally of depression (r = −0.15, p = 0.059, N = 163). Across disorders, tobacco adjustment was significantly correlated with odds ratios in non-tobacco studies (r = −0.19, p < 0.001, N = 366) and cannabis adjustment was significantly correlated with odds ratios in non-cannabis studies (r = −0.13, p = 0.010, N = 392). The indicator for adjustment for diet quality and/or sedentary behavior was nowhere significant.

In exploratory linear regression analyses, the set of study quality variables (sample size, specialty sample, disordered behavior, and control variables) explained 2.7% of the variance in odds ratios in the overall dataset and 12.4% of the variance in longitudinal studies. For cross-sectional studies of individual behaviors, regression models explained between 9.0% (tobacco use) and 27.5% (poor diet quality) of the variance in odds ratios, and for individual disorders between 1.4% (personality disorder) and 37.5% (bipolar disorder). See Supplementary Appendix B for the specification of these models. In these models, disordered behavior commonly reached significance as a positive influence on odds ratios, indicating that studies investigating such behaviors tended to find higher risk for mental disorders. In the full dataset, higher sample size reduced odds ratios, indicating that large-N studies tended to obtain more moderate risks; this was an issue especially for longitudinal (behavior before disorder) studies. In studies of cannabis and tobacco use, adjustment for its counterpart was a highly significant negative influence on odds ratios.

Findings for individual disorders

Psychosis

This review included 45 cross-sectional and 16 longitudinal figures for the association between unhealthy behaviors and psychosis (Tables 1a, 1b). The cross-sectional studies provide convincing evidence for associations between psychosis and cannabis and tobacco use and some evidence for associations with sedentary behavior and diet quality. The longitudinal studies provide some evidence that cannabis and tobacco use may lead to subsequent psychosis and more tenuous evidence that psychosis may lead to subsequent cannabis and tobacco use.

Depression

The present review included 208 cross-sectional and 43 longitudinal figures for the association between unhealthy behaviors and depression (Tables 2a, 2b). The cross-sectional studies provide convincing evidence for associations between depression and cannabis, tobacco, sedentary behavior, and diet quality. The longitudinal studies provide some evidence that sedentary behavior and use of cannabis and tobacco may lead to subsequent depression and that depression may lead to subsequent cannabis and tobacco use. There was very limited longitudinal evidence related to diet quality.

Anxiety

The review included 93 cross-sectional and 19 longitudinal figures for the association between unhealthy behaviors and anxiety (Tables 3a, 3b). The cross-sectional studies provide convincing evidence for associations between anxiety and cannabis, tobacco, sedentary behavior, and diet quality. The longitudinal studies provide some evidence that sedentary behavior and use of cannabis and tobacco may lead to subsequent anxiety and tenuous evidence that anxiety may lead to subsequent sedentary behavior and cannabis and tobacco use. No longitudinal evidence related to diet quality was included in this review.

Bipolar disorder

This review confirms that there is limited research on the association between unhealthy behaviors and bipolar disorder. A total of 25 cross-sectional and 10 longitudinal figures were included in the review (Tables 4a, 4b), many of which were based on the same NESARC dataset. The cross-sectional studies provide convincing evidence for associations between bipolar disorder and cannabis and tobacco use and tenuous evidence for associations with sedentary behavior and diet quality. The longitudinal studies provide some evidence that the use of cannabis and tobacco may lead to subsequent bipolar disorder, but no evidence that bipolar disorder may lead to subsequent cannabis and tobacco use. No longitudinal evidence related to sedentary behavior and diet quality was included in this review.