Marijuana use and pneumonia risk in a cohort of HIV-infected and HIV-uninfected men

Abstract

Background:

The prevalence of marijuana use is increasing in the United States. Marijuana smoking has been shown to impair the microbicidal activity of alveolar macrophages and decrease the number of ciliated epithelial cells in the bronchi with a parallel increase in the number of mucus-secreting surface epithelial cells, which may increase the risk of pneumonia. However, it remains unclear whether there is an association between smoking marijuana and pneumonia.

Methods:

Using data from the Multicenter AIDS Cohort Study (MACS), a long-term observational cohort study of men who have sex with men in the United States, we used Cox proportional hazards models to estimate the risk of pneumonia among HIV-infected (n = 2784) and HIV-uninfected (n = 2665) men from 1984 to 2013, adjusted for time-varying and fixed baseline covariates.

Results:

Weekly or daily marijuana use was not significantly associated with increased risk of pneumonia among HIV-uninfected men (adjusted hazard ratio; 95% confidence limits: 0.83, 0.56–1.23). In the disaggregated dose–response analysis, daily use (0.68, 0.34–1.35) was associated with a lower point estimate than weekly use [0.99, 0.79–1.25].

Conclusion:

Marijuana smoking was not associated with a significant increase in risk of pneumonia among HIV-infected or HIV-uninfected men.

Background

The prevalence of marijuana use in the United States has risen sharply in recent years, doubling in a 10-year span to 12.9% of the adult population in 2015 [1]. As of October 1, 2019, 33 states had medical marijuana laws, with 11 of those having additional provisions for recreational use [2]. As these trends continue, it is imperative to understand the full range of health risks associated with marijuana use.

One potential target for disease risk from marijuana use is the respiratory tract. Lower respiratory infections (pneumonias) are the third leading cause of disability adjusted life-years in the United States [3]. Pneumonia occurs when infectious agents successfully colonize the lower respiratory tract and cause inflammation contributing to significant mortality in children [4], the elderly [5], and hospitalized [6] populations. Pneumonia imposes a tremendous burden in the form of health care costs, as well as a social and economic burden on the working adult population [7]. Subclinical respiratory tract infections in healthy persons may also serve as a reservoir of infection in vulnerable and immunocompromised populations [8].

Tobacco smoking is the strongest risk factor for pneumonia followed by age, male sex, injection drug use, heavy alcohol use, hospitalization, and low CD4 cell count [9,10]. As marijuana is the second most commonly smoked substance, being typically delivered to the blood stream by combustion and inhalation, there is biological plausibility for a causal relationship between marijuana use and pneumonia [11–13].The introduction of volatile combustion products into the respiratory tract can promote low-grade inflammation increasing susceptibility to infection [14]. In addition, the production process of marijuana is largely unregulated making the leaves a potential source of pathogenic bacteria [15] and fungi [16]. Moreover, the sharing of paraphernalia could serve as a vehicle for the transmission of infectious agents. Marijuana also contains delta-9 tetrahydrocannabinol (THC), a molecular component not found in tobacco, which has immunosuppressive properties, mediated by cannabinoid-type 2 (CB2) receptors expressed on immune cells, including the inhibition of alveolar macrophage bactericidal and fungicidal activity [17]. Immunosuppression by THC may play a role in disease susceptibility, especially in an HIV-infected population. THC has been linked to increased viral replication in HIV-infected immunocompetent mouse models [18].

Previous observational studies have yielded conflicting or inconclusive results regarding the risk of pneumonia posed by marijuana [18–24]. One recent analysis of the Multicenter AIDS Cohort Study (MACS) found that marijuana use was associated with a modest increase in infectious and noninfectious pulmonary disease among HIV-infected men, but not among HIV-uninfected men [22]. Using the same data source but modified methodology and outcome definitions, we aimed to investigate the effects of marijuana smoking on risk of pneumonia in both HIV-infected and HIV-uninfected men.

Methods

Study cohort

The MACS is an ongoing multicenter prospective observational study established in 1984 of men who have sex with men. The study was designed to better understand the risk factors for and the natural and treated clinical history of HIV infection at four sites in the United States: Los Angeles, Chicago, Pittsburgh, and Baltimore/Washington D.C. [25] The rationale and study protocols have been described in detail [26]. Briefly, participants undergo standardized interviews detailing substance use, behavioral characteristics, medical treatments or conditions, a physical examination, and collection of biological specimens at semiannual study visits. All behavioral covariates were measured at these clinic visits by audio computer assisted self-interview (ACASI), a method with demonstrated accuracy in assessing sensitive behaviors [27,28]. All participants provided written informed consent, and the study protocols were approved by each site’s institutional review board. Detailed information about the study, including data collection forms, may be found at https://statepi.jhsph.edu/macs/macs.html.

Study population

From the initial 6821 participants available for analysis, we excluded those with demonstrated serological conversion to HIV infected status during follow-up (n = 636), as well as participants not contributing sufficient person-time within the risk period specified for each of three risk-sets (n = 736). This resulted in 2665 HIV-uninfected and 2784 HIV-infected participants who were eligible for our study. HIV-uninfected men were assessed for any pneumonia beginning in 1990, whereas HIV-infected men composed two risk-sets beginning in 1983: Pneumocystis carinii pneumonia (PCP) and non-PCP.

Exposure assessment

Marijuana use was the exposure of interest. Participants’ self-reported use of marijuana as “daily,” “weekly,” “monthly,” or “less often” since their last study visit. Marijuana use was also recategorized as a dichotomous variable to compare those who smoked at least weekly to those who smoked less than weekly.

Outcome assessment

Incident pneumonias were identified by medical record abstraction (ICD-9 codes 480–488) with the method of diagnosis being recorded, as available. In addition to medical record abstraction, pneumonia became independently reportable in the MACS in 1990. Participants were asked whether they had been diagnosed with pneumonia by a doctor or medical practitioner since their last visit. As the sensitivity to detect pneumonia by medical record abstraction alone among the HIV-uninfected participants was very low, we designated January 1990 (when independent reporting was introduced) as the baseline for pneumonia risk in the HIV-uninfected group. As we suspected a unique etiology and risk profile for PCP in contrast to all other types of pneumonia (non-PCP), we estimated risks for these outcomes independently.

Covariate assessment

We divided covariates into two categories: fixed at baseline and time-varying. Covariates fixed at baseline included sociodemographic variables continuous age, education (never attended college, attended at least some college, and attended graduate school), a composite of race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and other), and study site. Time-varying risk behavior covariates included tobacco smoking status since last visit (never, former, and current), frequency of alcohol use (daily, weekly, monthly, and never), inhalation of other illicit substances since last visit (cocaine, methamphetamine, or poppers), recent injection drug use (IDU), and history of IDU. Vaccination for streptococcal pneumonia (Pneumovax) since the last study visit was obtained by self-report. Hepatitis C virus infection status was determined using antibody and plasma viral RNA levels. Tuberculosis infection was obtained by self-report and confirmed by medical record review. CD4+ T cell count (cells/μL) were measured using flow cytometry [29]. HIV status was indicated by positive ELISA and confirmed by Western blot. Trained study coordinators and staff were available to answer questions, assist participants with items requiring clarification, and review responses for completeness during all study visits.

Statistical analyses

For cohort descriptive statistics, we used baseline covariate information (except for a subset of infrequently-occurring comorbidities, which we summarized across all visits). Cox proportional hazard models were used to assess the association between marijuana smoking and incident pneumonia stratified by HIV status. All time-varying covariates were lagged by one visit to establish clear temporal sequencing with respect to the outcome. HIV-infected participants could be censored by a competing risk if a different type of pneumonia occurred first. Participants became eligible to accrue time at-risk on study enrollment if they were free of pneumonia at baseline and could exit the cohort only by experiencing the outcome or being administratively censored by competing pneumonia, death, loss-to-follow-up, or the study end date of December 31st, 2013, whichever occurred first.

The final model was based on variables identified a priori in the literature as known risk factors for pneumonia. These included smoking status, education, age, race, recent hospitalization, history of pneumococcal vaccine, history of injection drug use, recent illicit drug use, and recent binge drinking. The covariates age and CD4+ cells/μL were treated as continuous variables. In the multivariable analysis, marijuana use was examined both in its original frequency of use and dichotomous forms. In addition, the variables calendar period and study site were hypothesized to be potential confounders in this cohort and were included based on their association with the outcome in the data (p < 0.2). We examined the interaction between tobacco smoking and marijuana use in a sensitivity analysis. All analyses were conducted using SAS v9.4 (SAS Institute, Cary, NC).

Results

Table 1 provides descriptive statistics for the study population at baseline and cumulative values for select characteristics. The mean age of the study population at baseline was 38 years (standard deviation [SD] = 8.6, range = 18.6–75.9). Most study participants (73%) were of non-Hispanic White race/ethnicity, recruited before 2001 (79%), and had completed some postsecondary education (81%). There was a balanced distribution of tobacco smoking status with 34% having never smoked, 29% having quit smoking, and 37% reporting current tobacco smoking at baseline. The most commonly reported frequency of marijuana use was monthly (51%) followed by none (27%), weekly use (15%), and daily use (7%).

Table 1.

Characteristics of study population at baseline (^* or last eligible visit), stratified by HIV-status

Sociodemographic, fixed
	Total	HIV-uninfected	HIV-infected
	(n = 5449)	(n = 2665)	(n = 2784)
Age (y)	%	%	%
Minimum	18.6	18.6	18.9
Mean	37.7	40.3	35.2
Std Dev	8.6	8.7	7.6
Maximum	75.9	75.9	70.0
Race
White	72.6	78.2	67.2
Black	17.7	14.7	20.5
Hispanic	5.9	4.1	7.5
Other	3.9	2.9	4.9
Center
Baltimore	24.1	26.9	21.5
Chicago	24.4	21.9	25.0
Pittsburgh	23.1	27.4	19.1
Los Angeles	29.3	23.9	34.4
Enrollment cohort
Pre-2001	78.6	81.8	75.4
Post-2001	21.4	18.2	24.6
Education
No college	19.1	14.9	23.1
Some college	50.6	48.4	52.8
Some graduate	30.3	36.7	24.1
Modifiable exposures, time-varying
	Total	HIV-uninfected	HIV-infected
	(n = 5449)	(n = 2665)	(n = 2784)
Tobacco smoking	%	%	%
Never	34.3	33.4	35.8
Former	28.9	35.1	19.1
Current	36.7	31.6	43.5
Marijuana
Never	27.2	36.7	18.2
Monthly	51.1	50.2	52.0
Weekly	14.5	8.7	20.0
Daily	7.1	4.4	9.7
Binge drinking	10.2	8.3	11.9
Drug use
Heroin (IDU)	4.1	3.5	4.6
Cocaine	22.6	21.7	43.0
Methamphetamines	16.9	7.7	25.5
Poppers	37.7	22.6	52.2
Any illicit (-IDU)	47.8	30.4	64.4
Comorbidities (ever*)
HCV	7.1	3.4	10.5
Hospitalized	41.9	41.4	42.4
Pneumococcal vaccine	26.4	9.6	42.3
Tuberculosis	0.7	0.6	0.9
CD4+ cell/mL
400+	84.1	99.1	69.7
200–399	11.3	0.8	21.4
100–199	2.9	0.1	5.5
0–99	1.6	0	3.2
Missing	0.2	0	0.3

^*Assessed at last visit.

Substance use (of all kinds) was more common in HIV-infected participants than in HIV-uninfected participants. Comorbidities are summarized based on tabulation across all visits with the presence at any visit being indicated by the percentages in Table 2. For example, by the time of last eligible study visit, 42% of all participants had experienced hospitalization at some point during the course of follow-up. The incidence rate of non-PCP in the HIV-uninfected and HIV-infected populations was 10.7 and 20.7 cases per 1000 person years, respectively. Outcome statistics for the study population, including total person-time, causal agent, X-ray confirmation, deaths, loss-to-follow-up and competing risks are provided in supplemental Table S1.

Table 2.

Univariate hazard ratios and 95% confidence limits for non-PCP pneumonia

Characteristic	HIV-uninfected			HIV-infected
	HR	LCL	UCL	HR	LCL	UCL
Marijuana use
Never	1.00	Ref	Ref	1.00	Ref	Ref
Monthly	0.95	0.74	1.21	1.18	0.94	1.48
Weekly	0.85	0.54	1.35	1.04	0.76	1.43
Daily	0.64	0.32	1.25	1.31	0.95	1.81
Race
White	1.00	Ref	Ref	1.00	Ref	Ref
Hispanic	1.13	0.66	1.94	0.78	0.54	1.13
Black	0.62	0.41	0.95	1.17	0.94	1.45
Other	0.90	0.42	1.90	0.56	0.32	0.98
Age
18–29 y	1.00	Ref	Ref	1.00	Ref	Ref
30–39 y	0.92	0.45	1.87	1.13	0.73	1.76
40–49 y	1.35	0.68	2.68	1.28	0.82	2.00
50–64 y	1.55	0.77	3.13	0.93	0.56	1.53
65+ y	1.46	0.62	3.43	1.48	0.65	3.34
Education
No college	1.00	Ref	Ref	1.00	Ref	Ref
Some college	0.83	0.58	1.19	0.86	0.69	1.08
Graduate	1.05	0.73	1.51	0.83	0.64	1.07
Study period
1983–1989	n/a	n/a	n/a	1.00	Ref	Ref
1990–1994	1.00	Ref	Ref	2.58	1.81	3.68
1995–1999	0.17	0.08	0.36	1.28	0.81	2.04
2000–2004	0.07	0.03	0.16	2.22	1.43	3.45
2005–2013	0.32	0.19	0.57	1.24	0.86	1.79
Study site
Baltimore	1.00	Ref	Ref	1.00	Ref	Ref
Los Angeles	0.59	0.44	0.79	1.05	0.83	1.32
Chicago	0.34	0.22	0.51	0.90	0.70	1.17
Pittsburgh	0.56	0.42	0.76	0.75	0.56	1.01
Tobacco smoking
Never	1.00	Ref	Ref	1.00	Ref	Ref
Current	1.26	0.92	1.72	1.56	1.25	1.95
Former	1.17	0.88	1.55	0.98	0.77	1.25
Single exposures
No	1.00	Ref	Ref	1.00	Ref	Ref
Binge drinking	0.80	0.33	1.94	0.99	0.46	2.15
Illicit drug use	0.91	0.69	1.19	0.98	0.82	1.18
IDU (recent)	1.08	0.35	3.35	2.91	1.71	4.96
IDU (ever)	1.66	1.06	2.59	1.62	1.19	2.20
Hospitalization	2.04	1.37	3.02	2.20	1.73	2.80
PV (ever)	2.22	1.57	3.13	1.11	0.91	1.35
CD4+ cell/mL
400+	—	—	—	1.00	Ref	Ref
200–399	—	—	—	1.94	1.57	2.39
100–199	—	—	—	2.39	1.76	3.25
0–99	—	—	—	4.73	3.67	6.10

HR = hazard ratio; IDU = injection drug use; LCL = 95% lower confidence level; PCP = Pneumocystis carinii pneumonia; PV = pneumococcal vaccine; UCL = 95% upper confidence level.

Table 2 provides the univariate hazard ratios and 95% confidence limits for each covariate level. An increase in the frequency of marijuana use was associated with decreasing point estimates for pneumonia risk among the HIV-uninfected population, with the lowest hazard ratio observed comparing daily to never users (Hazard Ratio = 0.64, 95% Confidence Interval: 0.32–1.25). In the HIV-infected group, the point estimate for daily use was above the null for non-PCP risk (HR = 1.31, CI: 0.95–1.81) and for PCP risk (HR = 1.24, CI: 0.85–1.81). Period effects were observed for all three risk sets, with a majority of cases ascertained before 1995. Low CD4+ cell count in the HIV-infected participants and past hospitalization in both the HIV-infected and HIV-uninfected participants were consistently associated with non-PCP risk. Tobacco smoking, ever injection drug use, and increasing age were associated with point estimates above the null for non-PCP, but these were not associated with PCP risk (supplemental Table S2). Point estimates for binge alcohol consumption were below the null for both types of pneumonia. Univariate hazard ratios for PCP among the HIV-infected group showed positive associations only with low CD4 counts and hospitalization (Table S2).

The final multivariable model was used to estimate the associations between pneumonia and marijuana, while adjusting for covariates identified in the univariate models (Table 3). The point estimate for the adjusted hazard of non-PCP for weekly (or more frequent) marijuana use was lower than the hazard ratio for the less frequent use (monthly or less) in the HIV-uninfected risk group (HR = 0.83, 95% CI 0.56–1.23). The point estimate for the adjusted association between weekly marijuana use and non-PCP in the HIV-infected group was near the null (aHR = 0.99, 95% CI: 0.79–1.25) while the estimate for weekly use and PCP was just above the null (aHR = 1.07, 95% CI: 0.88–1.29). The dose–response analysis revealed a consistently negative monotonic trend in point estimates for increasing marijuana use for non-PCP risk in the HIV-uninfected participants. Conversely, there was a positive trend in the point estimates for PCP risk with increasing marijuana use in the HIV-infected participants, while the estimates for non-PCP risk in the HIV-infected participants were equivocal, fluctuating around the null.

Table 3.

Adjusted^* hazard for PCP and non-PCP pneumonia in HIV-uninfected and HIV-infected participants, according to marijuana smoking frequency

Marijuana frequency	HR	LCL	UCL	Marijuana Frequency	HR	LCL	UCL
Population: HIV-uninfected; outcome: Non-PCP pneumonia
Never	1.00	Ref	Ref	Monthly (or less†)	1.00	Ref	Ref
Monthly	0.94	0.72	1.22	Weekly (or more)	0.83	0.56	1.23
Weekly	0.87	0.54	1.40
Daily	0.68	0.34	1.35
Population: HIV-infected; outcome: non-PCP pneumonia
Never	1.00	Ref	Ref	Monthly (or less†)	1.00	Ref	Ref
Monthly	1.08	0.84	1.37	Weekly (or more)	0.99	0.79	1.25
Weekly	0.93	0.66	1.30
Daily	1.10	0.78	1.56
Population: HIV-infected; outcome: PCP pneumonia
Never	1.00	Ref	Ref	Monthly (or less†)	1.00	Ref	Ref
Monthly	1.12	0.81	1.55	Weekly (or more)	1.07	0.88	1.29
Weekly	1.12	0.77	1.62
Daily	1.32	0.88	1.99

HR = hazard ratio; LCL = lower confidence limit; PCP = Pneumocystis carinii pneumonia; UCL = upper confidence limit.

^*Adjusted for confounding by study site, education, calendar period, age, race, recent hospitalization, history of pneumococcal vaccine, smoking status, injection drug use, illicit drug use, and binge drinking.

^†or less includes never.

Discussion

Data from the MACS from 1984 to 2013 adjusting for time-varying and fixed covariates (including confounding by study site, education, calendar period, age, race, recent hospitalization, history of pneumococcal vaccine, smoking status, injection drug use, illicit drug use, and binge drinking) does not provide evidence of a significant association between marijuana use and pneumonia incidence. A modest increase in the hazard of PCP among the immunocompromised group during the pre-HAART era was observed, although this has little relevance to the general population or the HIV-infected population living in the post-HAART era.

The primary strengths of this study include the repeated use of a structured and validated questionnaire along with clinical assessments at regular six-month intervals over a 30-year span in a large, well-characterized cohort of both immunocompetent and immunocompromised men. This provided us with statistical power to detect relatively modest associations for an outcome that occurs as infrequently as once per 100 years of cohort follow-up time, while adjusting for updated time-varying covariate data. However, the number of events observed over this period did limit our power to only detect associations with hazard ratios greater than approximately 1.40 and we were unable to exclude clinically meaningful associations below this value.

By relying on an observational cohort, the study is subject to several inherent limitations. First, the long duration of follow-up introduces period effects that could relate to risk. We addressed some of these analytically by adjusting for calendar time in a categorical fashion. With the advent of HAART in 1995, PCP risk was virtually eliminated. Non-PCP was ascertained adequately in the medical record abstraction for HIV-infected participants at the beginning of follow-up, but inadequately for the HIV-uninfected group until a question was introduced in 1990 to allow for self-report of pneumonia. If self-reported pneumonia was misclassified independently and nondifferentially with respect to marijuana use, the estimates would be biased toward the null. We expect a high degree of reliability as people are unlikely to misrepresent a medical diagnosis over a relatively short interval. Moreover, incidence rates for pneumonia demonstrated by the self-reporting variable were comparable to the estimated incidence rates reported in the literature [7]. As under-reporting of the exposure is possible since possession of marijuana was illegal at the time, the temporal ordering would preclude the exposure assessment from being dependent upon the outcome. The questionnaire did not assess mode of marijuana consumption; however, noncombustible options such as vaping are more recent innovations and combustion remains the primary mode [30]. Misclassification of the exposure could similarly bias the results toward the null. The MACS was not able to elicit income information from most participants, which is related to pneumonia incidence [31]. However, adjustment for education level was intended to address potential confounding by socioeconomic status.

If censoring by competing risks or loss to follow-up is related to pneumonia risk, the resulting selection bias would be small due to the relatively low frequency of these events. Dropout rates were low [32] although a substantial number of participants were removed from the risk-set by death over the long course of follow-up (n = 1670). The effect estimates can be interpreted as conditional risks (i.e., conditioned on death and loss to follow-up being independent of pneumonia risk). Censoring due to loss to follow-up (dropping out of the study, failure to continue study visits), could also contribute a potential selection bias to the effect estimates.

By relying on a population of men who have sex with men, the results may not generalize to other populations such as men who do not have sex with men or to women. Nonetheless, the results and conclusions for the HIV-uninfected group are potentially informative and relevant to the general population. Although PCP is no longer common in the post-HAART era, the HIV-infected group allows for investigation into the potential for THC to compound existing immunosuppression. The modest increase in the adjusted hazard for PCP is suggestive of enhanced immunosuppression warranting further investigation.

In the same study population, using slightly different outcome definitions, inclusion criteria and adjustment methods, Lorenz et al. [22] reached similar conclusions. When examining infectious pulmonary disease risk, they found an estimated aHR = 0.82 (95% CI: 0.57–1.20) among HIV-uninfected participants, and an aHR = 1.42 (95% CI: 1.09–1.86) among HIV-infected participants. Interestingly, they chose not to stratify or distinguish between the AIDS-defining opportunistic infection PCP and other types of pneumonia. Their decision to exclude all visits in which the CD4 + count was <200 cells/μL would have selectively removed many pneumonia outcomes and may have in turn induced an unpredictable selection bias as participants could enter and leave the risk set over the course of follow-up. Another difference in their analytic approach was to exclude all visits before 1996 and only adjust for period effects with a single cutoff at the year 2000. Despite the differences in approach, our results concur with theirs with regard to pneumonia risk in the HIV-uninfected group. Namely, that the risk interval is most consistent with no strong harmful effect of marijuana use on infectious pulmonary outcomes (their outcome), including pneumonia (our outcome), in the absence of HIV-infection. We found that the positive univariate association between pneumonia and marijuana use in the HIV-infected group was dampened on adjustment and only applied to PCP risk. These results support the hypothesis that the increase in pulmonary risk identified by Lorenz et al. may be due to an immunosuppressive effect of THC only in the absence of a fully competent immune system and with respect to opportunistic infections.

Using health information derived from a large electronic health system database Winhusen and colleagues [33], recently reported that cannabis-using patients with cannabis use disorder had a significantly higher incidence of pneumonia than control non-cannabis-using patients matched on demographic and socioeconomic variables with or without a positive history of tobacco-use disorder. As tobacco smoking is a well-established risk factor for pneumonia [34], it is possible that the authors’ failure to control for concomitant tobacco use, irrespective of a diagnosis of tobacco use disorder, contributed to the observed association.

We believe these results comprise important evidence regarding the relationship between marijuana use and lung health. Acknowledging the limitations of observational research, we hope these findings will promote further investigation into potential health risks posed by marijuana use.