If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Requests for reprints should be addressed to Paris Charilaou, MD, Department of Internal Medicine, Rutgers-Robert Wood Johnson Medical School/Saint Peter's University Hospital, 254 Easton Ave, New Brunswick, NJ 08901.
The nationwide prevalence of cannabis use/abuse has more than doubled from 2002 to 2011. Whether the outpatient trend is reflected in the inpatient setting is unknown. We examined the prevalence and incidence of cannabis abuse/dependence as determined by discharge coding in a 10-year (2002-2011) National Inpatient Sample, as well as various trends among demographics, comorbidities, and hospitalization outcomes.
Methods
Cannabis abuse/dependence was identified on the basis of International Classification of Diseases, 9th Revision, Clinical Modification codes 304.3* and 305.2* in adults aged 18 years or more. We excluded cases coded “in remission.” National estimates of trends and matched-regression analyses were conducted.
Results
Overall, 2,833,567 (0.91%) admissions with documented cannabis abuse/dependence were identified, patients had a mean age of 35.12 ± 0.06 years, 62% were male, and there was an increasing trend in prevalence from 0.52% to 1.34% (P <.001). The mean Charlson Comorbidity Index was 0.47 ± 0.006, and inpatient mortality was 0.41%. All of the above demonstrated an increasing trend (P <.001). Mean length of stay was 6.23 ± 0.06 days. The top primary discharge diagnoses were schizoaffective/mood disorders, followed by psychotic disorders and alcoholism. Asthma prevalence in nontobacco smokers had a steeper increase in the cannabis subgroup than in the noncannabis subgroup (P = .002). Among acute pancreatitis admissions, cannabis abusers had a shorter length of stay (−11%) and lower hospitalization costs (−7%) than nonabusers.
Conclusion
Cannabis abuse/dependence is on the rise in the inpatient population, with an increasing trend toward older and sicker patients with increasing rates of moderate to severe disability. Psychiatric disorders and alcoholism are the main associated primary conditions. Cannabis abuse is associated with increased asthma incidence in nontobacco smokers and decreased hospital resource use in acute pancreatitis admissions.
Cannabis abuse/dependence prevalence in the inpatient increased from 0.5% to 1.3%.
•
Abuse/dependence has an increasing trend in older, sicker patients.
•
The most common coded comorbidities are psychiatric disease and alcoholism.
•
Cannabis abuse is associated with an increase in asthma numbers in nontobacco smokers.
•
Cannabis abuse is associated with shorter length of stay (−11%) and hospital costs (−7%) in acute pancreatitis admissions.
Cannabis has been a highly debated and controversial issue nationwide. It is the most prevalent illicit substance in the United States and the second most widely smoked substance after tobacco, ranging from simple one-off recreational use to abuse and overt dependence with associated psychiatric illnesses and socioeconomic implications for individuals.
Indications for its medicinal use and legalization vary from state to state, potentially influencing the geographic distribution of cannabis and its abuse. By 2016, 25 states had already passed medical marijuana laws, 16 of which were enacted between 1996 and 2011.
Although there is scarce evidence of the prevalence of cannabis abuse and related medical comorbidities in the adult inpatient population, its association with psychiatric diseases is well known.
The overall prevalence of cannabis use across the nation has more than doubled from 2002 (4.1%) to 2012 (9.5%), with up to a 31.3% prevalence among those aged 18 to 34 years. In addition, cannabis use disorder has increased to 2.9% in 2012 from 1.5% in 2001.
Whether the cannabis use disorder trend is reflected in the inpatient setting is not known.
In this analysis, we examine the prevalence of cannabis abuse over 10 years (2002-2011) among inpatient admissions using the Nationwide Inpatient Sample (NIS). We studied the trends and association of cannabis with various demographic factors, clinically related comorbidities, and hospital-associated outcomes, such as disability on discharge, length of stay, inpatient mortality, and costs.
Materials and Methods
Design and Data Source
We performed a retrospective observational cohort study, using the Healthcare Cost and Utilization Project NIS database, sponsored by the Agency for Healthcare Research and Quality (AHRQ).
The Healthcare Cost and Utilization Project NIS is the largest publicly available database in the United States and includes a 20% sample of all inpatient admissions from 1000 hospitals in more than 40 states. This sample averages 7 million discharges every year, representing more than 95% of the US population.
We used the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes to identify discharges with cannabis abuse/dependence–coded diagnoses between 2002 and 2011.
Study Population
Within the inclusion criteria were all adults aged 18 years or more who had a diagnosis of cannabis abuse or dependence (excluding remission) in any of the discharge diagnosis fields (DX1-DX25). The cannabis abuse/dependence diagnoses (“continuous,” “episodic,” “unspecified”) were identified using the following ICD-9-CM codes, excluding those “in remission”: 304.30, 304.31, 304.32, 305.20, 305.21, and 305.22.
For trend analysis, observations with missing data in age, gender, or inpatient mortality were excluded. Similar methodology was used in previous studies of chronic obstructive pulmonary disease/bronchiectasis, asthma, and tobacco smoking.
The total number of admissions with cannabis abuse/dependence over the period 2002 to 2011 was 2,833,567 (0.91% of total admissions), of which 0.6% was coded as the primary discharge diagnosis. The proportion of all admissions involving cannabis abuse/dependence had an increasing trend (P <.001) from 0.52% (2002) to 1.34% (2011), representing a 260% increase over 10 years, along with a concurrent increasing incidence from 737 per million of total US population in 2002 to 1773 per million in 2011 (Figure 1). Cannabis dependence disorder overall comprised 17.3% of the cannabis subpopulation.
Figure 1Admissions with cannabis abuse/dependence trends overview. Left Y-axis: number of admissions with 95% CIs from population estimates. Right Y-axis: admissions per million of population (grey, continuous line), hospitalization costs (grey, dotted line) with 95% CIs from population estimates.
The mean age of cannabis abusers (35.12 ± 0.06 years) has increased by 3.2 years (P <.001). This is due to both decreasing trends in younger age groups (ages <50 years) and increasing trends in older groups (Table 1). Men were the predominant gender (62.5%) with an increasing trend (P <.001). Differences among races showed weaker linear trends, with whites (55.2%) and Hispanics having a decreasing trend, and blacks and Asians showing an increasing trend (P <.001). Native Americans had a strong linear increasing trend (P <.001), with a 300% increase (0.5% in 2002 to 1.5% in 2011). In terms of median income levels, the lowest and highest quartiles showed an increasing percentage of patients discharged with a cannabis abuse/dependence diagnosis, with middle-income quartiles steadily decreasing (P <.001).
Table 1Main Trends in Cannabis Abuse/Dependence Disorder in the Inpatient Population
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
0.52% ± 0.04%
0.57% ± 0.03%
0.73% ± 0.03%
0.75% ± 0.04%
0.86% ± 0.04%
0.95% ± 0.05%
0.95% ± 0.06%
1.11% ± 0.05%
1.25% ± 0.07%
1.34% ± 0.07%
0.91% ± 0.02%
<.001
Increasing
Admissions incidence per million of population every year
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
737 ± 161
809 ± 169
1028 ± 222
1053 ± 219
1212 ± 260
1324 ± 292
1320 ± 302
1512 ± 330
1661 ± 391
1773 ± 397
-
<.001
Increasing
Cannabis dependence
23.0%
18.0%
19.6%
18.7%
18.3%
16.7%
16.1%
16.5%
15.8%
15.4%
17.3%
<.001
Decreasing
Cannabis abuse as primary discharge diagnosis
0.69%
0.36%
0.58%
0.48%
0.43%
0.41%
0.38%
0.38%
0.43%
0.24%
0.4%
<.001
Decreasing
Cannabis dependence as Primary discharge diagnosis
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
AHRQ = Agency for Healthcare Research and Quality; AIDS = acquired immunodeficiency syndrome; COPD = chronic obstructive pulmonary disease; N/A = not available; PTSD = post-traumatic stress disorder; SE = standard error.
∗ Weak linear trend (linear regression P >.05).
† Cannabis abuse/dependence admissions and % from total admissions, incidence per million of population, age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
‡ These comorbidities were extracted per the ICD-9-CM diagnoses codes as outlined in Supplementary Table 1 (available online).
Discharges with moderate-to-severe disability made up 12.4% of total cannabis abuse/dependence admissions over the 10-year period and showed a weak linear increasing trend (P <.001); further trends within this subset are described in Supplementary Table 3 (available online).
Mean length of stay (excluding inpatient deaths) was 5.23 ± 0.06 days and had a weak decreasing linear trend, whereas hospitalization costs have been increasing (P <.001), averaging $7514 ± $114. Medicare and Medicaid as payers (49.6% combined) showed an increasing trend (P <.001).
Clinically relevant, associated, chronic conditions were assessed for trends using AHRQ comorbidity indicators. Alcoholism (28.1%), chronic lung disease (13.8%), psychoses (10.7%), and depression (10.2%) constituted the majority of these comorbidities, with alcoholism showing a decreasing trend (decreased by 10.7%, P <.001), whereas depression and psychoses showed an increasing trend (P <.001). Post-traumatic stress disorder increased from 2.59% to 4.18% over 10 years (P <.001), with an overall prevalence of 3.38%. Of note, chronic lung disease
had a 1.7 times increase in the period of 10 years, with a strong increasing trend (P <.001). Other chronic conditions such as congestive heart failure, renal failure, pulmonary hypertension, and metastatic cancer also showed increasing trends (P <.001). The Charlson Comorbidity Index followed in concordance with these findings, almost doubling during the study period (0.313 in 2002 vs 0.600 in 2011; P <.001).
The 20 most common primary discharge diagnoses among the patients who had cannabis abuse/dependence as a secondary diagnosis are outlined in Supplementary Figure (available online). More than 5500 unique primary discharge diagnoses (ICD-9-CM codes) were recorded in this subpopulation of patients. As shown in Supplementary Figure (available online), approximately 1 of 5 patients (19.3%) was admitted for a primary diagnosis of mood/schizoaffective disorder. Psychosis and other schizophrenia accounted for 5.7% of the primary discharge diagnoses, followed by drug-induced mental disorders (3.5%), alcohol-induced mental disorders (2.2%), and opioid dependence (1.5%). Acute pancreatitis (ICD-9-CM: 577.0) showed an increasing trend from 2002, entering the Top 20 list in 2007 (1.13%), continuing its incremental trend (P <.001), and reaching position number 9 in 2011 (1.75%). Overall, acute pancreatitis was the primary discharge diagnosis for 1.18% of all admissions with documented cannabis abuse/dependence. Given this finding, we conducted a secondary analysis of the cholelithiasis prevalence trend (ICD-9-CM: 574.0-574.9)
in the cannabis abuse/dependence subpopulation and found a doubling in prevalence (0.4% in 2002 vs 0.9% in 2011; P <.001), with the trend having a similar slope as that of acute pancreatitis (0.00052 vs 0.00050; slope Δ P = .816). Furthermore, we tested the hypothesis that among all patients with a primary discharge diagnosis of acute pancreatitis, cannabis abuse/dependence could be more common among the chronic pancreatitis users. The findings were consistent with our hypothesis that cannabis use was 58% more likely in patients with chronic pancreatitis than in patients without the disease (odds ratio [OR], 1.58; 95% confidence interval [CI], 1.48-1.69; P <.001), after adjusting for Charlson Comorbidity Index, gender, age, alcoholism, depression, psychiatric disease, and metastatic cancer. Of note, the prevalence of chronic pancreatitis in the cannabis abuse/dependence population more than doubled over the study period (0.51% in 2002 to 1.05% in 2011, P <.001).
Cannabis and Prevalence Trends of Respiratory Disease
After exact matching (outlined in the Online Supplement), linear regression comparing the slopes of trends in prevalence of asthma and chronic obstructive pulmonary disease separately between cannabis and noncannabis users was done. There were no significant differences in chronic obstructive pulmonary disease trend prevalence (Figure 2A). However, in asthma trend prevalence, the nonsmokers group showed a significant difference in slopes (1.7 times steeper increase, P = .002) (Figure 2B), whereas the significant difference was lost (P = .230) (Figure 2B) in the tobacco smokers subgroup, suggesting that cannabis has an effect on asthma prevalence only among nontobacco smokers.
Figure 2Bivariate comparison among cannabis and tobacco smoking in lung disease trends. (A) Age- and gender-matched chronic obstructive pulmonary disease/bronchiectasis prevalence trends. (B) Chronic obstructive pulmonary disease/bronchiectasis prevalence trends are plotted for each year, and a linear regression line is fitted for each subgroup. There was no significant difference between the slopes of cannabis versus noncannabis users in both the tobacco and nontobacco groups. (B) Age- and gender-matched asthma prevalence trends. (A) Asthma prevalence trends are plotted for each year, and a linear regression line is fitted for each subgroup. Among nontobacco smokers, cannabis users had a steeper increase of asthma prevalence than noncannabis users. Among tobacco smokers, there was no significant difference between cannabis and noncannabis users.
Matched regression analysis between cannabis and noncannabis abuse/dependence users was conducted separately for asthma and acute pancreatitis as the primary discharge diagnoses (Table 2). In the asthma model, length of stay was slightly decreased by 6% (time ratio, 0.94; 95% CI, 0.894-0.996; P = .036), whereas costs were not affected (OR, 1.01; 95%, CI, 0.96-1.06; P = .815) by cannabis abuse/dependence after correcting for alcoholism, tobacco smoking, race, and hospital-level variables. Among the acute pancreatitis admissions, both length of stay (time ratio, 0.89; 95% CI, 0.84-0.94; P <.001) and costs (OR, 0.93; 95% CI, 0.88-0.97; P = .003) were significantly decreased in the cannabis abuse/dependence group, while also adjusting for history of chronic pancreatitis.
Table 2Effect of Cannabis Abuse/Dependence on Hospital Outcomes
Primary Discharge Diagnosis
Length of Stay
Hospitalization Costs
Time Ratio
95% CI
P Value
OR
95% CI
P Value
Asthma
0.944
0.894-0.996
.036
1.01
0.96-1.06
.815
Chronic obstructive pulmonary disease
0.98
0.90-1.07
.615
1.05
0.97-1.13
.271
Acute pancreatitis
0.89
0.84-0.94
<.001
0.93
0.88-0.97
.003
Controls were matched on exact age, exact Charlson comorbidity index, gender, specific hospital, and year. Average number of matched controls per cannabis abuse/dependence patient: 3.34. Additional adjustment by multivariable regression was done for tobacco smoking, white race, alcoholism, hospital bed size, hospital region, urban hospital location, and hospital teaching status. For the acute pancreatitis model, adjustment for chronic pancreatitis was made. Note: A time ratio <1 is interpreted as an earlier discharge (shorter length of stay) and vice versa. The OR for hospitalization costs indicates percent change in the presence of cannabis abuse/dependence compared with its absence.
Our analysis demonstrated an approximately 2-fold increase in the prevalence of cannabis abuse/dependence-related admissions, with increasing trends in older age groups and their associated comorbidities and increasing hospitalization costs. Furthermore, a substantial proportion of these admissions are related primarily to psychiatric disease (including substance abuse), as previously described.
In addition, there was a trend toward abuse/dependence in both the poorest and most wealthy quartiles of the study population. Of note, the data also reveal an increased prevalence of acute pancreatitis as a primary discharge diagnosis over this 10-year period. Of note, nontobacco smokers who abuse cannabis showed a steeper incidence of asthma compared with noncannabis users. Finally, cannabis abusers being admitted for acute pancreatitis are independently associated with shorter length of stay and lower hospitalization costs.
A clearly increasing trend of admissions involving cannabis abuse/dependence is evident. This could be explained partly from the legalization,
Medical marijuana laws in 50 states: investigating the relationship between state legalization of medical marijuana and marijuana use, abuse and dependence.
to various extents, of cannabis as a medical or recreational substance, which in turn increases the prevalence of users in the general population and by extension the chance to observe them in the inpatient setting. This finding correlates with the results from a recent outpatient interview study.
We observed that cannabis users who are admitted to the hospital are becoming older, accompanied by a larger comorbidity burden as evident by the Charlson Comorbidity Index score and various other chronic AHRQ comorbidities, a finding that is expected with increasing age. Because patients who are admitted are increasingly more clinically complicated, we can see a correlating increasing trend of hospitalization costs and a weakly increasing trend of moderate-to-severe disability-on-discharge rates over time, with ages >50 years tripling in percentage within this discharge disability subgroup. Medical use of cannabis is potentially another explanation for the aging population of inpatient cannabis users. Legalization of medical cannabis over the last 15 years has led to extended research and subsequently its use in a wider spectrum of medicine, such as chemotherapy-associated nausea and vomiting, multiple sclerosis, neuropathic pain, acquired immunodeficiency syndrome–related anorexia and cachexia, and advanced cancer pain.
Indeed, the latter 2 have shown increasing trends in our analysis. Because these conditions are more likely to be seen in patients aged >50 years, this could explain the doubling of the 50- to 69-year age group over the last 10 years. Since there is no way of capturing “medical cannabis use” in the NIS while differentiating it from abuse and dependence, and given the simultaneous increase of conditions that cannabis is indicated as an adjunctive therapy,
Such a link could be explored further in patients with these conditions who have been prescribed medicinal cannabis by following them for development of abuse/dependence over a period of years.
Moreover, a socioeconomic/racial picture seems to emerge, with white men maintaining predominance, and the poor and the rich with increasing abuse/dependence versus the middle-class. A significantly increasing trend (tripling in 10 years) in the proportion of the Native American minority could reflect an overall increase in their outpatient use of cannabis. The West region has the lowest percent of cannabis abusers, even though the state of California legalized cannabis in 1996, 6 years before the beginning of our study period. Future studies that examine possible confounders, such as prevailing age and comorbidity difference between geographic sections, may help explain this finding.
Approximately 1 of 3 inpatient cannabis users have a history of alcoholism. It is the most common AHRQ comorbidity we analyzed, but, in fact, this is decreasing over the years. Its popularity with cannabis use has been a known social issue
Trends in marijuana and other illicit drug use among college students: results from 4 Harvard School of Public Health College Alcohol Study surveys: 1993-2001.
; however, there is no clear reason why there is a decreasing trend. This could be due to a possible “substitution” effect, with cannabis replacing alcohol, among other substances.
The increasing incidence of acute pancreatitis admissions within the cannabis abuse/dependence inpatient subpopulation was an unforeseen finding, considering the fact that alcoholism (a common cause of acute pancreatitis) has a decreasing trend. In an effort to evaluate another possible cause for the increasing trend of acute pancreatitis, we secondarily examined the trend of gallstone disease as a possible cause of pancreatitis. We discovered a significant doubling over 10 years (0.4%-0.9%) and a similar trend slope. Furthermore, it seems that an increasing proportion of patients among the cannabis group have chronic pancreatitis, which could explain the use of cannabis as a pain relief adjunct. In fact, patients with acute-on-chronic pancreatitis are 58% more likely to abuse cannabis, compared with those with nonchronic pancreatitis. Whether this increase can be explained by cannabis use being associated with higher body mass index and subsequently a higher chance for gallstone disease,
Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013.
is unknown and would nevertheless merit further analysis in future studies.
Among the admissions of patients with asthma, we observed a nonclinically significant, yet statistically significant decrease of 6% in length of stay in patients with cannabis/dependence abuse. We expect this finding to be secondary to increased power of the study, rather than a true observation, especially considering the nonsignificant effect on hospitalization costs (Table 2). Similar findings are shown for chronic obstructive pulmonary disease admissions. On the contrary, among the acute pancreatitis admissions, results were more striking: Length of stay was significantly shorter (−11%), and hospitalization costs were significantly less (−7%) in patients with cannabis abuse/dependence, even after adjusting for history of chronic pancreatitis. We hypothesize 2 possible explanations for this finding: First, the half-life of cannabis in the plasma in chronic users is approximately 4 days
; therefore, this finding could be explained by the chronic use of cannabis, resulting in a more seamless hospital stay with easier pain control. Furthermore, we hypothesize that chronic cannabis users would rather try and go home earlier, by subjectively reporting less pain, to use their cannabis at home for further pain relief. Nevertheless, this finding would merit of further research in the future.
Our study exhibits evidence of the association between cannabis use and increasing asthma prevalence among nontobacco smokers from the general population. Among tobacco smokers, the cannabis abuse/dependence did not have any significant contribution to increasing asthma prevalence. This suggests that tobacco smoking has a strong enough effect that any cannabis effect would be masked, although it was shown in other studies that cannabis and tobacco have an additive synergy in predisposing asthmatic attacks.
Cannabis did not seem to affect the prevalence trend of chronic obstructive pulmonary disease, regardless of tobacco smoking status. This could be secondary to tobacco smoking being a more chronic influence on a chronic disease such as chronic obstructive pulmonary disease, because tobacco has been readily available to the masses for a longer time than cannabis.
Study Limitations
The use of administrative databases has been increasing in clinical research, but as with any such database, they are prone to coding inaccuracies and deficiencies, and mainly rely on the clinician's documentation and the coder's experience.
Specifically, although the cannabis ICD-9-CM codes were used previously in the literature, it has been shown that, in general, illicit drug abuse/dependence is suffering from under-reporting (ie, low sensitivity), with the positive predictive value ranging from 48% to 65%.
Validation of key behaviourally based mental health diagnoses in administrative data: suicide attempt, alcohol abuse, illicit drug abuse and tobacco use.
This has been attributed to the fact that coding for these diagnoses was occurring only when there was documentation of the clinician advising the patient for cessation or if the patient reported problem behaviors (which also explains the high prevalence of mood disorders in our study).
Validation of key behaviourally based mental health diagnoses in administrative data: suicide attempt, alcohol abuse, illicit drug abuse and tobacco use.
Naturally, it is expected that cannabis abuse would be under-reported by patients because of its controversial social and legal nature, making it difficult to accurately extrapolate the findings of this study to the outpatient setting and more generally to the whole population.
The NIS does not provide data on a patient's last use of cannabis before admission; therefore, to increase our specificity, we excluded cases that were coded as “in remission.”
Conclusions
Cannabis abuse/dependence is on the rise in the inpatient population, with older and sicker patients becoming the majority with increased rates of moderate to severe disability on discharge. The inpatient cannabis subpopulation follows specific socioeconomic profiles and is mainly associated with mood disorders, whereas there is a signal of a possible relationship with disorders related to obesity, such as acute pancreatitis and gallstone disease. Cannabis abuse may be contributing to increased asthma incidence in nontobacco smokers. Finally, cannabis abuse is associated with decreased hospital resource use in acute pancreatitis admissions. Further studies that analyze subpopulations with these diseases between cannabis and noncannabis users are warranted.
Appendix: Online Supplement
Variables
The Nationwide Inpatient Sample (NIS) variables were used to identify demographic information, including age, gender, and race. Hospitalization-related variables included primary payer (insurance types), bed size, region, teaching versus nonteaching, urban versus rural, and median household income for patient ZIP code (categorized by quartile from 2003 onwards). Select Agency for Healthcare Research and Quality comorbidity variables
from the NIS were extracted and analyzed for trends. The variables of chronic obstructive pulmonary disease/bronchiectasis, asthma, tobacco smoking, and post-traumatic stress disorder were identified according to International Classification of Diseases, 9th Revision, Clinical Modification codes outlined in Supplementary Table 1 (available online). A modified version of the Charlson Comorbidity Index was calculated for every observation
Gordon M. Calculates comorbidity indices based on ICD-9/10 (comorbidities.icd10). 0.6.1 ed. Available at: https://github.com/gforge/comorbidities.icd10 2016. Accessed June 1, 2016.
Length of stay was estimated after excluding patients who died in the hospital. Total charges billed for hospital services per hospital stay were converted to costs using the cost-to-charge tables provided by the Healthcare Cost and Utilization Project (HCUP), weighted for missing data and adjusted for annual inflation.
Disability-on-discharge severity was computed using the “DISPUniform” variable, as described in Supplementary Table 3 (available online).
Statistical Analysis
Categoric and continuous variables were compared using chi-square and t test/Mann-Whitney tests, where appropriate, respectively. Tests for trend included the Cochran-Armitage test
Signorell A. DescTools: Tools for Descriptive Statistics. Vol R package version 0.99.16 2016. Available at: https://cran.r-project.org/package=DescTools. Accessed March 21, 2016.
in R and linear regression to assess the linearity of proportions for categoric variables, as well as the Cuzick nonparametric test in Stata (StataCorp LP, College Station, Tex) for continuous variables.
Categoric variables with P >.05 in linear regression of proportions against year were considered “weak linear trends,” even when the Cochran-Armitage test was significant for trend.
Incidence of hospitalizations with cannabis abuse/dependence was calculated per million US population, by dividing the weighted national estimates of hospitalizations in a given year by the US census population estimate ≥18 years of age for the corresponding year.
Population estimates - vintage 2009: annual estimates of the resident population by sex and selected age groups for the United States: April 1, 2000 to July 1, 2009 (NC-EST2009-02).
Population estimates - vintage 2011: annual estimates of the resident population by sex and selected age groups for the United States: April 1, 2010 to July 1, 2011 (NC-EST2011-02).
The outcomes of length of stay and hospitalization costs were calculated with lognormal accelerated failure time model and a linear regression model, respectively. For length of stay, the event was defined as variable DIED = 0, meaning that the patient was discharged, whereas DIED = 1 implying in-hospital death was the censored observation. In this way, the survival model would allow to retain power without upfront exclusion of deaths from the analysis. The accelerated failure time model also is not subject to the proportional hazards assumption, in contrast to the Cox proportional hazards model. The time ratio calculated in this accelerated failure time model is essentially a percentage change of length of stay (ie, time ratio = 1.5 means 50% increase in length of stay). For hospitalization costs, given their exponential distribution, they were first log-transformed and the coefficients of the model exponentiated to produce odds ratios (ie, odds ratio = 1.5 means 50% increase in cost). The analysis for both outcomes was stratified by exact matching on gender, exact age in years, exact Charlson Comorbidity Index, and year-hospital-id. The matched controls (drawn from a 15% random sample of noncannabis abuse/dependence patients) in each matching stratum were then weighted as the inverse of the total controls in a stratum, essentially creating arithmetically 1:1-matched clusters. Each matching stratum corresponds to a cluster of matched cases, which was used to correct the regression coefficients' standard errors for using matched observations. Race information was collapsed to white versus nonwhite race to enhance power. Smoking and race were then included as additional confounders in the regression models, together with other hospital-level variables. This analysis was done in Stata 14, using the matching module “cem”.
Likewise, a 15% random sample of noncannabis abuse/dependence patients were matched on exact age, gender, year, and tobacco smoking status with cannabis abuse/dependence patients, and the annual prevalence trends of chronic obstructive pulmonary disease/bronchiectasis and asthma were compared between cannabis and noncannabis users. A linear regression line was fitted on the prevalence (% cases) of the selected variable against years. The slopes (β term) of the fitted lines were then compared for significant difference. The latter analysis was done in GraphPad Prism 6.0f (GraphPad Software, La Jolla, Calif).
All analyses used national estimates that were calculated using the TRENDWT NIS variable, as suggested by the HCUP.
Calculation of standard errors for population estimates was done by taking into account possible missing hospitals among the cannabis subpopulation and the complex survey design of the NIS, by using the “SVY” commands in Stata (except for matched analyses), as recommended by the HCUP.
P values were a priori designated significant as <.025 for trends and <.05 for bivariate and other comparisons.
Supplementary FigureTop 20 primary discharge diagnoses in cannabis abuse/dependence. Graphical illustration of the distribution of various, arbitrarily categorized, primary discharge diagnoses (DX1) over the years 2002 to 2011. Purple: mood disorder excluding schizophrenia/psychoses. Green: alcoholism-related disorders or any drug withdrawal–related disorders (excluding opioids). Yellow: opioid abuse/dependence. Red: schizophrenia-related disorders. Blue: unspecified psychoses. Turquoise: acute Pancreatitis. White: chest pain, pneumonias, or any other conditions not listed.
The DISPUNIFORM variable was used to calculate moderate-to-severe disability on discharge cases after excluding inpatient deaths and “against medical advice” discharges. Values 2, 5, and 6 were considered “moderate-severe discharge,” which include transfers to short-term hospital, skilled nursing facility, intermediate care facility, another type of facility, and home health care. Note: Hospital deaths are excluded in table.
AHRQ = Agency for Healthcare Research and Quality; AIDS = acquired immunodeficiency syndrome; COPD = chronic obstructive pulmonary disease.
∗ Weak linear trend (linear regression P >.05).
† Age, length of stay, hospitalization costs, and Charlson Comorbidity Index are presented with their linearized standard errors population estimates.
‡ These comorbidities were extracted according to the ICD-9-CM diagnoses codes as outlined in Supplementary Table 1 (available online).
Medical marijuana laws in 50 states: investigating the relationship between state legalization of medical marijuana and marijuana use, abuse and dependence.
Trends in marijuana and other illicit drug use among college students: results from 4 Harvard School of Public Health College Alcohol Study surveys: 1993-2001.
Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013.
Validation of key behaviourally based mental health diagnoses in administrative data: suicide attempt, alcohol abuse, illicit drug abuse and tobacco use.
30094-3&id=gr1.jpg){kind=link}
30094-3&id=gr2.jpg){kind=link}
30094-3&id=fx1.jpg){kind=link}