This study was conducted in Songkhla, southern Thailand, which is located on the coast of the Gulf of Thailand. The total population of the province is approximately 1.4 million people, with a population density of 194 per square kilometer (41). In Songkhla, the two largest hospitals responsible for TB treatment were Songklanagarind Hospital and Hatyai Hospital. The former, a university hospital, offers super-tertiary care, whereas the latter is classified as a regional hospital operated by the Ministry of Public Health. Thus, both hospitals were selected as the study hospitals. Regarding cannabis consumption, the Thailand National Poll reported a 32% prevalence of lifetime exposure to any type of cannabis. However, its prevalence at the provincial level has not been reported (40). Furthermore, this study was conducted in 2023, when both medicinal and recreational uses of cannabis were legal.

A matched case–control study was conducted to compare cannabis usage and smoking between patients with active pulmonary TB and healthy controls in the same population.

The case group was defined as active pulmonary TB patients. Patients were recruited from among outpatients diagnosed with active pulmonary TB based on bacteriological results or clinical diagnoses. Inclusion criteria comprised patients over 20 years old, residents of Songkhla Province, and those undergoing treatment at the study hospitals since January 1st, 2023. The latter criterion was placed to make sure that TB cases were diagnosed after cannabis was legalized in 2022. We excluded TB patients with diabetes mellitus, cancer, and HIV infection to prevent unnecessary heterogeneity in the exposure groups. The nursing teams at both study hospitals identified eligible cases for recruitment process. Information on the actual residential addresses of the patients who participated in the study was obtained for planning community control selection.

Controls in this study was matched with the cases using age group, sex, and community as the matching variables. Control matching varied between urban and rural cases due to differences in TB transmission, influenced by factors such as high population density, crowded living conditions, and low hygiene (42, 43). For rural areas, we defined a community as a village in which the patient resided. For the urban setting, we defined the study community as an alley of the patient’s home. The responsible healthcare center for each study community was approached. The study protocol was explained to the patients. The health offices were requested to invite health volunteers in charge to participate in the control selection. Information on the age group (10 year range) and sex of the patients in that community was used to match with the cases and guide the health volunteers to select controls. The exclusion criteria for controls were the same as those for the case group. Potential controls were not included if they had been diagnosed with TB. However, we did not attempt to exclude patients with TB from the control group using any tests, including chest radiography. When more than one eligible control was available for each TB case, volunteers were advised to randomly select one. Selected controls were then interviewed at a nearby community health center.

This study sample size was calculated using two independent proportion formulas. For the exposure proportion in the control group, we used the previously reported prevalence of Thai people who had ever used cannabis for medicinal purposes, which was 0.32 for P₁. We assumed the proportion of cannabis consumption among the cases (P₂) to be 0.49, which was sufficient to generate an OR of 2. With a significance level of 0.05, a power of 0.8, and a 1:1 case-to-control ratio, the required total sample size was 148 cases and 148 controls.

We started recruiting participants on March 9th, 2023, and ended data collection on December 5th, 2023. Both the case and control groups provided informed consent. Participants were informed that this was a survey on cannabis use without disclosing our hypothesis of the link to TB. Face-to-face interviews were conducted using a structured questionnaire. The questionnaire contained three sections including demographic information, tobacco and alcohol consumption, and cannabis consumption.

The outcome variable was active pulmonary TB status, as described in the aforementioned protocol. Our main exposure variables were cannabis exposure, including oral consumption and smoking status. Oral cannabis consumption included drinking cannabis tea, consuming foods enhanced or mixed with cannabis derivatives, and consuming commercial drinks containing cannabis. Smoking details included the form of preparation/instrument used for the smoking session, such as rolled joint, mixed cigarette, pipe, and bong. In this study, a bong was classified as any type of water pipe used for smoking cannabis. The social aspects of smoking included whether cannabis was shared during the session, such as sharing a cannabis joint or bong. As we hypothesized that cannabis sharing would be the main mechanism of TB transmission among cannabis users, we stratified smoking cannabis levels into non-smoking, smoking, and smoking and sharing cannabis separately for both smoking via a bong and rolled joints. Timing variables of consumption included age at start and quit (among ex-users). Frequencies and reasons for consumption were also collected. Other potential confounding variables included a history of tobacco smoking and alcohol consumption, education, occupation, and monthly income.

During the interview process, data were collected directly and entered into a mobile REDCap application (44). The mobile data were transferred and securely stored in the REDCap database hosted by the Research and Development Office of Prince of Songkla University (45, 46). Data from REDCap were exported in a comma-separated value format (CSV), specifically for the analysis process, using the R program version 4.2.3 (47).

Descriptive statistics were used to summarize all independent variables and stratified by the outcome variable. Frequency and percentage were used as categorical variables, whereas continuous variables were described using mean and standard deviation (SD) or median and interquartile range (IQR) based on their distribution. Univariate association analysis was conducted using either the chi-square test or Fisher’s exact test for small expected cell counts, while the Student’s t-test or Mann–Whitney U test was performed for continuous variables depending on the variable’s distribution. A logistic regression analysis was performed to adjust for the remaining confounding factors. The key independent variables maintained in the model were cannabis use and smoking. Other covariates included in the model as potential confounders were those that showed significantly association with the outcome in the univariate analysis (p < 0.2).

Because smoking tobacco is a strong confounder, a multivariate model was initially performed without it. This model was then compared with the model in which it was included. The coefficient/OR of cannabis in the model, including smoking tobacco, reflected its direct effect on TB risk. This stepwise modeling approach was used separately to analyze the effects of cannabis use and cannabis smoking. To compare the different models, we used the Akaike information criterion (AIC) values. The model with the lowest AIC was considered the best-fit model, given that the set of records was the same. In the case of collinearity between two exposure variables, the variable with the lowest AIC was selected. This model was then summarized to display the OR with a 95% confidence interval (CI) for each independent variable. In addition, the PAF based on Levin’s formula was computed for modifiable risk factors, particularly cannabis use (48). The significance level (α) for multivariate analysis was set at 0.05.

Of 152 eligible TB patients, 148 patients gave their consent, a response rate of 97.4%. We obtained 117 age-sex-community matched controls from the health volunteer recruitment. As these controls were readily recruited by the health workers, we did not have the information on the number of eligible subjects and the response rate. Table 1 shows the results of the univariate analysis. The patients in the case group were generally younger than the controls, had a lower education level, and had a higher proportion of unemployed individuals and office workers. Substance usage was more common in the case group, with a significantly higher proportion of individuals reporting ever-drank alcohol and tobacco, along with a higher prevalence of current smoker. Additionally, the monthly cigarette dosage was significantly higher in those who had smoked tobacco.

The details of cannabis consumption among cases and controls are summarized in Table 2. Eleven percent of the controls were current cannabis users, whereas nearly 19% had used cannabis at least once before, most commonly through oral consumption (12.8%). Cannabis consumption among cases and controls was not significantly different among current users, quitting age, usage frequency, oral consumption, and health reasons behind cannabis use. Cases were more likely to have a history of ever using cannabis and smoking cannabis via a bong or joint, especially when sharing cannabis during the session. We obtained the ventilation status from 37 out of 38 bong smokers and 30 out of 33 joint smokers. Smoking in well-ventilated areas was reported by 33 (89.2%) of bong smokers and 26 (86.7%) of joint smokers.

Table 3 compares the AICs of all prediction models for active pulmonary TB from various types of cannabis use with different combinations of covariates. Owing to the collinearity between smoking via the bong and joints, they were separately analyzed and displayed. Row A displays the results from ever-used cannabis, while rows B and C display those from smoking cannabis via bong and joint, respectively. Model 1, incorporating all exposure variables, showed a significant crude increased risk that remained significant after age, sex, education and occupation were included (Model 2). Of all the models, model 3B yielded the lowest AIC value and was therefore used to calculate the ORs and population-attributed fractions.

Table 4 displays the ORs (95% CI) of variables in model 3B. The effects from smoking and sharing cannabis via a bong were independently significant, as were the effects from tobacco smoking. The levels of ORs between smoking and sharing cannabis via a bong and tobacco smoking were similar (4.22; 95% CI: 1.47–12.07 and 4.36; 95% CI: 1.97–9.65, respectively). The percentages of PAF were 12.16 and 12.62 for the respective categories. Based on the likelihood ratio tests, all other independent variables in the model, except age, were also statistically significant.