Association of bidi smoking and tobacco-use patterns with cancer risk in India: a multicentre unmatched case-control study (2024–2025)

Summary

Background

Despite widespread use of bidi and other tobacco products in India, multicentre studies examining product-specific risks for cancer remain limited. This study aimed to determine association of bidi smoking and tobacco-use patterns (exclusive, dual, and poly-tobacco use) in relation to cancer risk, metastatic presentation and treatment modalities.

Methods

This hospital-based unmatched case-control study was conducted across eight tertiary-care centres in seven Indian states (2024–2025). Cases were recruited based on histologically confirmed cancer and controls were selected based on no prior history of cancer. Individuals with severe illness, or inability to provide reliable information were excluded. Associations were assessed using Firth-penalised logistic regression and adjusted for confounders.

Findings

Among 4500 screened, 2037 cases (1143 females, 894 males) and 2066 controls (553 females, 1513 males) were enrolled. Tobacco use was reported by 608 cases (29.8%) and 364 controls (17.6%). Compared with non-tobacco users, exclusive bidi smokers (82 cases, 30 controls) had three times higher odds (adjusted Odds Ratio [aOR] 3.11 [95% CI 1.86–5.31]), smokeless tobacco users had (329 cases, 260 controls) 1.5 times higher odds (aOR 1.55 [95% CI 1.24–1.95]), dual bidi-smokeless users nearly four times higher odds (aOR 3.88 [95% CI 2.11–7.36]), and poly-tobacco users over four times higher odds of any cancer (aOR 4.54 [95% CI 2.65–7.88]). Bidi duration (≥10 years) (aOR 6.75 [4.43–10.47]) and number of bidis (≥10 bidis/day) (aOR 7.19 [4.57–11.51]) were also associated with increased cancer risk. Tobacco use was also significantly associated with metastatic presentation and treatment outcomes.

Interpretation

Bidi smoking and its combinations with other tobacco products showed strong associations with cancer risk, while other tobacco-use patterns also demonstrated significant associations; however, findings should be interpreted cautiously due to the observational design. These findings support the need for strengthened bidi regulation and targeted tobacco cessation policies.

Funding

This work was supported with funding from Bloomberg Philanthropies Bloomberg Initiative to Reduce Tobacco Use.

Research in context.

Evidence before this study

PubMed, Web of Science, and Google Scholar were searched from inception to February 2026, with no language restrictions, using the terms “bidi smoking”, “smokeless tobacco”, “poly-tobacco”, “tobacco use”, “cancer risk”, “cancer incidence”, “India”, “South Asia”, “case-control study”, and “cohort study”, used singly and in combination. Reference lists of included studies and relevant systematic reviews and meta-analyses were additionally reviewed. Studies were included if they reported quantitative associations between tobacco exposure and cancer outcomes in human populations. Although the carcinogenicity of smoked and smokeless tobacco is unequivocally established by the International Agency for Research on Cancer (IARC), existing evidence for bidi-specific cancer risk is largely confined to single-site analyses (particularly oral and lung cancers), individual Indian cohort studies with limited geographic scope, or studies that do not disaggregate risk by tobacco product type. A 2024 systematic review and meta-analysis (JCO Global Oncology) confirmed a significant association between tobacco use and cancer incidence in India but noted substantial heterogeneity and a paucity of product-specific, multi-cancer risk estimates. Comparative data quantifying differential cancer risk across exclusive bidi use, exclusive smokeless tobacco use, dual-use, and poly-tobacco use patterns and their relationship to metastatic presentation and oncologic treatment outcomes remain absent from the literature.

Added value of this study

This multi-centre, hospital-based unmatched case-control study, conducted across eight tertiary care centres from seven Indian states enrolled 2037 cancer cases and 2066 controls. It provides product-disaggregated analysis of tobacco-associated cancer risk in India. The study demonstrates that exclusive bidi smoking confers the highest cancer risk among all tobacco product categories (adjusted odds ratio 3·11, 95% CI 1·86–5·31), exceeding that of exclusive cigarette and exclusive smokeless tobacco use, with a clear exposure–response gradient established for both duration and daily frequency of bidi use. Dual and poly-tobacco use conferred markedly amplified risks, suggesting supra-additive carcinogenic effects of combined tobacco exposures. Beyond cancer occurrence, this study characterised the association between tobacco-use patterns with metastatic presentation and treatment modality.

Implications of all the available evidence

Taken together, existing global evidence and the findings of this study highlight that bidi smoking and poly-tobacco use represent a disproportionate cancer burden in India. The demonstrated dose–response relationship and synergistic risk with combined tobacco use reinforce the need for strengthened product-specific regulation of bidis, which currently benefits from substantially weaker regulatory oversight than cigarettes in India. National tobacco control programmes in India should integrate bidi-specific cessation counselling protocols and risk communication strategies that address the widespread misperception of bidis as a safer alternative to cigarettes. Clinicians managing cancer patients in India should elicit detailed poly-tobacco exposure histories, given the association with higher metastatic burden and altered treatment patterns observed in this study. Future research should prioritise prospective cohort designs with biochemical validation of tobacco exposure, investigation of gene-environment interactions, and population-based studies to confirm the generalisability of these risk estimates beyond tertiary-care settings.

Introduction

Tobacco products use constitute one of the leading preventable causes of death and disease worldwide, accounting for an estimated 7.69 million deaths and 200 million disability-adjusted life-years annually, with a well-established causal link to cancer.¹ The International Agency for Research on Cancer (IARC) has classified tobacco smoke and smokeless tobacco as Group 1 carcinogens, indicating sufficient evidence for their role in human carcinogenesis.² Bidi smoking, which are inexpensive hand-rolled tobacco products, is associated with elevated cancer risks, with hazard ratios of 3.55 for oral cancer and 5.54 for respiratory and intrathoracic cancers, exceeding those observed for cigarette smoking, and showing strong dose–response relationships across multiple malignancies, including cancers of the oral cavity, pharynx, lung, and other aerodigestive sites.^3,4 At the population level, bidi smoking continues to contribute to disease burden in India, with dose–response increases in all-cause mortality (risk ratios 1.3 for 1–7 bidis/day and 2.2 for ≥8 bidis/day) and a persistent prevalence of 7.7% among adults in 2016–2017 despite a 16.4% relative decline since 2009–2010, with marked socio-demographic and regional variation.^5,6 While age-standardised daily smoking prevalence has declined globally to 25.0% in men and 5.4% in women, representing reductions of 28.4% and 34.4% since 1990, tobacco use remains high and heterogeneous in low- and middle-income countries such as India, where diverse products sustain a persistent public health challenge.^7,8

The Indian subcontinent exhibits a unique and heterogeneous tobacco landscape, characterised by the concurrent use of smoked forms (e.g., bidi, cigarettes) and a wide variety of smokeless tobacco products (e.g., khaini, gutkha, zarda).⁹ These include chewing products (e.g., khaini, sun-dried tobacco with slaked lime), areca nut–tobacco mixtures (e.g., gutkha), and flavoured tobacco preparations (e.g., zarda), which are placed in the oral cavity without combustion.⁹ This diversity is compounded by important regional variations in consumption patterns, influenced by cultural practices, socio-economic status, and product availability.⁹ Bidis have been associated with a higher risk of oral, pharyngeal, and lung cancers compared to cigarettes, due to higher concentrations of specific toxins per gram of tobacco.⁵ In parallel, smokeless tobacco use remains highly prevalent in India and is associated with increased risks of oral cancer (relative risk 1.8–5.8) and oesophageal cancer (relative risk 2.1–3.2).^10,11 Smokeless tobacco use is associated with markedly increased risks of oral cancer (odds ratios 1.48–27.4) and pancreatic cancer (odds ratios 1.6–2.1), supporting a causal relationship.^2,12

Many existing studies have focused on individual tobacco products or broad categories, often failing to capture the real-world scenario of dual or poly-tobacco use, which is common in the Indian population.¹³ Furthermore, the differential impact of specific tobacco habits on cancer outcomes beyond incidence, such as the risk of metastasis at diagnosis or the requirement for more intensive treatment, is not well characterised. Characterising these associations is important for prognostication and for the rational allocation of resources within healthcare systems. Previous studies have reported strong associations between bidi smoking and upper aerodigestive tract cancers.^1,3,5,14,15 India has a heterogeneous tobacco-use landscape influenced by regional, socio-economic, and cultural differences. However, large-scale multi-site studies evaluating bidi use and its combinations with other tobacco products across diverse populations remain limited, restricting understanding of comparative product-specific risks and exposure–response relationships. To address these evidence gaps, the present study employed a multi-site, hospital-based case-control design across seven states in India. The objective of this study was to determine the association between bidi smoking and cancer risk, compare these associations across different tobacco-use patterns (exclusive, dual, and poly-tobacco use), examine dose–response relationships with bidi use, and assess associations with metastatic presentation and treatment modalities.

Methods

A multicentre, hospital-based, case-control study was conducted across eight tertiary care centres in seven states in India from March 2024 to March 2025. Participants were recruited from following centres: All India Institute of Medical Sciences (AIIMS) Rajkot, AIIMS Gorakhpur, AIIMS Rishikesh, AIIMS Guwahati, AIIMS Kalyani, AIIMS Patna, Gauhati Medical College and Hospital (GMCH), Guwahati, and Mahatma Gandhi Institute of Medical Sciences (MGIMS), Sevagram. These sites were selected to represent states with varying prevalence of tobacco use, as reported in the Global Adult Tobacco Survey (GATS-2, 2016–2017), ranging from 25.1% in Gujarat and 26.6% in Maharashtra to 33.5% in West Bengal, 35.5% in Uttar Pradesh, and 48.2% in Assam, thereby capturing regional variation in tobacco-use patterns across India.^16,17 [Supplementary Fig. S1]

Cases were defined as individuals diagnosed with various types of cancer, including but not limited to lung cancer, oral cancer, oesophageal cancer, stomach cancer, pancreatic cancer, and cervical cancer. Cases were identified from hospital records at the participating sites; diagnoses were based on histopathological confirmation or documented clinical diagnosis. In instances of multiple primary malignancies, cases were categorised based on the primary cancer diagnosis as documented in the medical records to ensure consistent outcome classification. The distribution of cancer types among cases is provided in Supplementary Table S1.

Controls were defined as individuals with no history of any cancer. Cancer-free status was ascertained through self-report and verified against available medical records at the time of recruitment. They were recruited from visitors or attendants of patients attending the same hospitals for non-neoplastic conditions. This source population was selected to ensure that controls arose from the same catchment population and had comparable healthcare access and utilisation patterns as cases, thereby minimising selection bias.

Eligible cases were identified from oncology clinics and inpatient records and approached consecutively at each centre; controls were approached from attendants or visitors accompanying patients with non-neoplastic conditions in the same hospitals during the study period. Individual matching was not performed to avoid potential overmatching and to retain the ability to examine the independent effects of key variables.¹⁸ Instead, cases and controls were recruited from the same hospitals and catchment populations to enhance comparability of the source population. Potential confounding was addressed through multivariable adjustment for key covariates.

Eligible cases were adults with a cancer diagnosis supported by medical records or histopathological confirmation. Patients with cancer who were deceased at the time of data collection were excluded. Controls were required to be free of any prior cancer diagnosis. Individuals with a history of cancer treatment, recurrence, inability to provide informed consent, or insufficient medical documentation were excluded from participation. Controls with medical conditions that could plausibly distort the association between tobacco use and cancer risk were excluded, including individuals with severe or terminal illness, chronic diseases strongly associated with tobacco use (e.g., advanced cardiovascular or chronic respiratory diseases), metabolic conditions such as diabetes and obesity, substance use disorders (e.g., alcohol dependence), neurological or cognitive impairment affecting recall (e.g., dementia or psychiatric illness), and those unable to provide reliable information.19, 20, 21, 22 These selection criteria were predefined and applied uniformly across all study sites.

The sample size was determined a priori using G∗Power (Version: 3.1.9.4) for an unmatched case–control design to detect an association between tobacco use and cancer occurrence. The exposure prevalence among controls was set at 28.6%, based on estimates from the GATS-2.²³ Previous epidemiological studies from India have reported moderate to high effect sizes for tobacco use and cancer risk, particularly for specific cancer sites and higher exposure levels.^3,15 Given the heterogeneity of tobacco-use patterns, inclusion of multiple cancer types, and multivariable adjustment for potential confounders, a conservative minimum detectable effect size was specified a priori to ensure adequate statistical power for detecting modest population-level associations. An odds ratio of 1.27 was specified for the sample size estimation, corresponding to an expected exposure prevalence of 33.0% among cases.^24,25 Using a two-sided significance level (α) of 0.05, statistical power of 90%, and a 1:1 case–control ratio, the required sample size was calculated to be 3638 participants (1819 cases and 1819 controls). To account for a potential 10% non-response or attrition rate, the sample size was adjusted to approximately 4043 participants. While the overall sample size provides estimates for the primary analysis, subgroup and secondary analyses, including metastatic presentation and treatment modalities, were considered exploratory.

Data were extracted and collected by trained interviewers using a structured questionnaire administered in the local language when participants visited the centre. The questionnaire was adapted from standardised instruments, including GATS and the World Health Organization STEPS survey.^26,27 Two trained interviewers were assigned per site; they followed standardised guidelines for participant approach, informed consent, and questionnaire administration, and were blinded to the specific analytical hypotheses. Sociodemographic information included age, biological sex, education, occupation, household income, religion, marital status, and dietary pattern. Behavioural data included a detailed history of tobacco use and current alcohol consumption. Exposure was defined as ever use prior to cancer diagnosis for cases and ever use prior to the six months preceding data collection for controls. This differential reference period was utilised to ensure temporal precedence and mitigate protopathic bias, preventing disease-induced behavioural modifications (reverse causality) from confounding the exposure assessment.^28,29 Exposure was categorised a priori into no tobacco use, exclusive bidi smoking, cigarette smoking, smokeless tobacco use, dual-use combinations, and poly-tobacco use involving three or more products, based on cumulative use patterns, established epidemiological evidence, and the biological plausibility of increasing exposure burden.^30,31 For bidi smokers, detailed quantitative exposure measures were collected as continuous variables, including duration of use (in years) and daily frequency (number of bidis per day). These variables were subsequently categorized into <10 and ≥10 years for duration, and <10 and ≥10 bidis per day for frequency. These exposure gradients were defined a priori to enable assessment of dose–response relationships. These variables were self-reported by participants. Dose–response analyses were specifically conducted for bidi smoking because bidi use represented the primary exposure and principal focus of the study.

Clinical information comprised height and weight for the calculation of Asian-specific BMI categories, along with physician-diagnosed hypertension and diabetes.³² Height and weight were measured directly by trained interviewers at each participating tertiary care centre using standardised procedures. For cancer cases, clinical details including cancer type, metastatic status at diagnosis, and treatment modalities were extracted from hospital records. Metastatic status was determined based on stage at the time of initial diagnosis, as recorded in medical records, and categorised as present or absent. Treatment modalities were operationalised based on the number of therapeutic modalities received and categorised as single treatment (any one modality alone: chemotherapy, surgery, radiation therapy, hormone therapy, immunotherapy, or other cancer-directed therapy), dual treatment (any two modalities in combination), multimodal treatment (three or more modalities), and supportive/palliative care. These variables were obtained from hospital records.

All data were subsequently transmitted to a central nodal site at AIIMS Gorakhpur for unified data entry, quality validation, and consistency checks to ensure uniform classification of exposure categories and outcomes across sites.

Measures were undertaken to minimise potential sources of bias. Selection bias was reduced by recruiting cases and controls from the same hospitals and catchment populations. Information bias was minimised using standardised questionnaires administered throughout the sites. Recall bias was addressed by using standard data collection and defined reference periods for exposure assessment. Reverse causation (protopathic bias) was minimised by assessing exposure prior to diagnosis among cases. Misclassification bias was reduced by confirming cancer diagnoses through histopathological reports or medical records. Potential confounding was addressed through a priori selection of covariates and multivariable adjustment. Inter-site variability was minimised through centralised data entry, quality validation, and consistency checks. Supplementary Fig. S1 illustrates the methodology of the study.

Statistical analysis

The primary outcome was cancer status (case/control). Secondary outcomes included metastatic status at diagnosis and treatment modalities. The primary exposure was tobacco use. Dose–response variables for bidi smoking included duration of use (years) and daily frequency (number of bidis per day). Covariates included age (years), biological sex, education, occupation, household income, religion, dietary pattern, body mass index, physician-diagnosed hypertension and diabetes, and current alcohol use.¹⁴

Statistical analyses were conducted using R version 4.4.1 within the RStudio interface (version 2026.01.0 Build 392). Fully annotated R scripts used for data processing, regression modelling, and statistical testing are provided in the Supplementary File. All tests were two-sided, with statistical significance set at p ≤ 0.05. Baseline characteristics were summarised using descriptive statistics, with continuous variables presented as mean ± Standard Deviation (SD) and categorical variables as frequencies and percentages. Covariate imbalance between cases and controls was assessed using Standardized Differences (Std. Diff).

Associations between tobacco-use patterns and cancer risk were examined using multivariable Firth-penalised logistic regression, with adjusted odds ratios (aORs) reported.³³ Firth-penalised regression reduces bias from sparse data and separation, yielding stable estimates.³³ Dose–response relationships for bidi smoking were assessed by modelling duration and daily frequency as ordered categorical variables in separate regression models. The Cochran–Armitage test was used for assessing the trend. Associations between tobacco-use patterns and metastatic status were evaluated using Pearson’s chi-square test, with effect size quantified using Cramér’s V and trend assessed using the Cochran–Armitage test.³⁴ Differences in treatment modalities across tobacco-use categories were analysed using Pearson’s chi-square test with Cramér’s V for effect size, and the Kruskal–Wallis test for ordinal comparisons. In the presence of sparse cell counts, Fisher’s exact test with Monte Carlo simulation was applied.

All regression models used a forced-entry approach, with covariates entered simultaneously based on biological plausibility and prior evidence.³⁵ Multicollinearity was assessed using the Generalized Variance Inflation Factor (GVIF). Adjusted GVIF values ranged from 1.02 to 1.07, indicating negligible multicollinearity [Supplementary Table S2]. Correlation matrices were examined to evaluate pairwise associations, and no strong correlations (r ≥ 0.70) were observed. Model adequacy was evaluated using penalised log-likelihood values, likelihood ratio statistics and Akaike Information Criterion [Supplementary Table S3]. Covariates and tobacco exposure categories were defined a priori based on established epidemiological evidence and biological plausibility, with exposure levels ordered to reflect increasing cumulative risk, thereby avoiding data-driven classification and potential post hoc bias.^29,35

Missing data were handled using complete-case analysis. For key variables (occupation and income), missingness was treated as an explicit category to retain observations and avoid introducing assumptions. Comparisons between participants with and without missing socioeconomic data demonstrated predominantly small-to-moderate imbalance, with the greatest differences observed for tobacco-use pattern and BMI category [Supplementary Table S4]. Measures were undertaken to ensure robustness of the findings, including assessment of multicollinearity and model adequacy using penalised likelihood-based statistics.³⁶ Sensitivity analyses were additionally performed by modelling bidi smoking duration and daily frequency as continuous variables to assess the robustness of the observed dose–response associations.

Ethics statement

The study protocol was reviewed and approved by the Institutional Ethics Committees of all participating centres: All India Institute of Medical Sciences (AIIMS) Rajkot (AIIMS.Rajkot/IEC/67/2023), AIIMS Gorakhpur (IHEC/AIIMS-GKP/BMR/264/2024), AIIMS Rishikesh (AIIMS/IEC/23/417), AIIMS Guwahati (AIIMSG/IEC/M4/F125/2024), AIIMS Kalyani (IEC/AIIMS/Kalyani/certificate/2024/149), AIIMS Patna (AIIMS/Pat/IEC/2023/115), Gauhati Medical College and Hospital, Guwahati (190/2007/Pt-51/Dec. 2023/1), and Mahatma Gandhi Institute of Medical Sciences, Sevagram (MGIMS/IEC/COMMED/255/2023). The study was conducted in accordance with the principles of the Declaration of Helsinki and applicable institutional guidelines. Written informed consent was obtained from all participants prior to enrolment in the study.

Role of the funding source

The funders had no role in the study design; collection, analysis, or interpretation of data; writing of the report; or in the decision to submit the article for publication.

Results

A total of 4500 individuals were screened for recruitment across eight tertiary-care centres in India, resulting in a final study population of 4103 participants (2037 cancer cases and 2066 controls) after exclusion criteria were applied. Gender-specific analysis are provided in Supplementary Table S5. Baseline characteristics indicated substantial heterogeneity between cases and controls [Table 1]. While the mean age was comparable (53.5 ± 13.19 years in cases vs. 52.4 ± 13.68 years in controls; Std. Diff = 0.083; p < 0.001), marked imbalances were observed in other sociodemographic domains.

Table 1.

Demographic and socioeconomic characteristics of participants.

Demographic variables	Control (n = 2066)	Cases (n = 2037)	Total (n = 4103)	Standardized difference	p - value
Age (Mean ± SD)	52.36 ± 13.68	53.48 ± 13.19	52.91 ± 13.45	0.083	0.008a
Age category
<40	480 (23.2)	373 (18.3)	853 (20.8%)	0.185	0.001a
40–50	520 (25.2)	457 (22.4)	977 (23.8%)
50–60	468 (22.7)	598 (29.4)	1066 (26.0%)
60–70	387 (18.7)	416 (20.4)	803 (19.6%)
>70	211 (10.2)	193 (9.5)	404 (9.8%)
Gender
Male	1513 (73.2)	894 (43.9)	2407 (58.7%)	0.455	0.001a
Female	553 (26.8)	1143 (56.1)	1696 (41.3%)
Education category
Illiterate	365 (17.7)	413 (20.3)	778 (19.0%)	0.360	0.001a
Primary School	351 (17.0)	389 (19.1)	740 (18.0%)
Middle School	302 (14.6)	314 (15.4)	616 (15.0%)
High School	432 (20.9)	367 (18.0)	799 (19.5%)
Intermediate/Diploma	232 (11.2)	380 (18.7)	612 (14.9%)
Graduate	307 (14.9)	151 (7.4)	458 (11.2%)
Professional Degree	77 (3.7)	23 (1.1)	100 (2.4%)
Occupation
Unemployed	292 (14.1)	194 (9.5)	486 (11.8%)	0.398	0.001a
Unskilled	273 (13.2)	187 (9.2)	460 (11.2%)
Semi-skilled	223 (10.8)	319 (15.7)	542 (13.2%)
Skilled	253 (12.2)	445 (21.8)	698 (17.0%)
Professional	14 (0.7)	27 (1.3)	41 (1.0%)
Housewife	443 (21.4)	709 (34.8)	1152 (28.1%)
Missing	568 (27.5)	156 (7.7)	724 (17.6%)
Income				0.564
<10,000	684 (33.1)	426 (20.9)	1110 (27.1%)		0.001a
10,000–20,000	463 (22.4)	503 (24.7)	966 (23.5%)
20,000–30,000	195 (9.4)	369 (18.1)	564 (13.7%)
30,000–50,000	117 (5.7)	316 (15.5)	433 (10.6%)
>50,000	69 (3.3)	191 (9.4)	260 (6.3%)
Missing	538 (26.0)	232 (11.4)	770 (18.8%)
Religion
Hindu	1833 (88.7)	1798 (88.3)	3631 (88.5%)	0.393	0.001a
Muslim	37 (1.8)	173 (8.5)	210 (5.1%)
Other	196 (9.5)	66 (3.2)	262 (6.4%)
Diet
Vegetarian	958 (46.4)	720 (35.3)	1678 (40.9%)	0.226	0.001a
Non-Vegetarian	1108 (53.6)	1317 (64.7)	2425 (59.1%)
Location – States
West Bengal	242 (11.7)	238 (11.7)	480 (11.7%)	0.010	1.000
Assam	401 (19.4)	392 (19.2)	793 (19.3%)
Uttar Pradesh	351 (17.0)	349 (17.1)	700 (17.1%)
Bihar	205 (9.9)	200 (9.8)	405 (9.9%)
Gujarat	152 (7.4)	147 (7.2)	299 (7.3%)
Uttarakhand	380 (18.4)	378 (18.6)	758 (18.5%)
Maharashtra	335 (16.2)	333 (16.3)	668 (16.3%)

Values are presented as mean ± standard deviation (SD) for continuous variables and frequency (n, %) for categorical variables. Age was analysed both as a continuous variable (years) and categorical. Standardized differences >0.10 indicate meaningful imbalance. Variables with standardized differences >0.10 were included as covariates in multivariable regression models. Continuous variables were compared using the independent t-test, while categorical variables were analysed using Pearson’s chi-square test. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

Tobacco use was reported by 608 (29.8%) of cancer cases compared with 364 (17.6%) of controls (Std. Diff = 0.337, p < 0.001). Among specific product categories, exclusive smokeless tobacco use was the most common in both groups (329 [16.2%] of cases vs. 260 [12.6%] of controls), followed by exclusive bidi smoking (82 [4.0%] vs. 30 [1.5%]). Poly-tobacco use was concentrated among cases; for instance, the combination of bidi and cigarette smoking was reported by 17 (0.8%) of cases compared with only 3 (0.1%) of controls [Table 2].

Table 2.

Distribution of clinical and behavioural risk factors, including bidi and other tobacco product use.

Category	Control (n = 2066)	Cases (n = 2037)	Total (n = 4103)	Standardized difference	p - value
Asian BMI category				0.521
Underweight	192 (9.3)	560 (27.5)	752 (18.3)		0.001a
Normal weight	871 (42.2)	809 (39.7)	1680 (40.9)
Overweight	680 (32.9)	506 (24.8)	1186 (28.9)
Obese	323 (15.6)	162 (8.0)	485 (11.8)
Diabetes				0.044
Present	193 (9.3)	217 (10.7)	410 (10.0)		0.161
Absent	1873 (90.7)	1820 (89.3)	3693 (90.0)
Hypertension
Present	350 (16.9)	291 (14.3)	641 (15.6)	0.073	0.019a
Absent	1716 (83.1)	1746 (85.7)	3462 (84.4)
Alcohol habit
Present	18 (0.9)	240 (11.8)	258 (6.3)	0.460	0.001a
Absent	2048 (99.1)	1797 (88.2)	3845 (93.7)
Tobacco type
No Habit	1702 (82.4)	1429 (70.2)	3131 (76.3)	0.337	0.001a
Bidi only	30 (1.5)	82 (4.0)	112 (2.7)
Smokeless tobacco only	260 (12.6)	329 (16.2)	589 (14.4)
Cigarettes only	13 (0.6)	30 (1.5)	43 (1.0)
Bidi and smokeless tobacco	19 (0.9)	59 (2.9)	78 (1.9%)
Bidi and cigarette	3 (0.1)	17 (0.8)	20 (0.5%)
Cigarette and smokeless tobacco	10 (0.5)	20 (1.0)	30 (0.7%)
Poly-tobacco use	29 (1.4)	71 (3.5)	100 (2.4%)
Duration of bidi use
Non-bidi users	1985 (96.1)	1808 (88.8)	3793 (92.4%)	0.301	0.001a
<10 years	42 (2.0)	60 (2.9)	102 (2.5%)
≥10 years	39 (1.9)	169 (8.3)	208 (5.1%)
Frequency of bidi use per day
Non-bidi users	1985 (96.1)	1808 (88.8)	3793 (92.4%)	0.291	0.001a
<10 per day	35 (1.7)	144 (7.1)	179 (4.4%)
≥10 per day	46 (2.2)	85 (4.2)	131 (3.2%)

Values are presented as frequency (n, %) unless otherwise specified. Body mass index (BMI) was classified using Asian-specific BMI categories: underweight (<18.5 kg/m²), normal weight (18.5–22.9 kg/m²), overweight (23.0–24.9 kg/m²), and obese (≥25 kg/m²). Standardised differences >0.10 indicate meaningful imbalance. Variables with standardized differences >0.10 were included as covariates in multivariable regression models. All categorical variables were analysed using Pearson’s chi-square test. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

Compared with non-users (reference category), exclusive bidi smoking was associated with approximately three times higher risk of cancer (aOR 3.11, 95% CI 1.86–5.31; p < 0.001). Exclusive smokeless tobacco use was also significantly associated with cancer (aOR 1.55, 1.24–1.95; p < 0.001). Dual use of bidi and smokeless tobacco was associated with nearly four times higher odds of cancer risk (aOR 3.88, 2.11–7.36; p < 0.001). Cigarette and smokeless tobacco dual use was also independently associated with increased risk (aOR 2.64, 1.10–6.58; p = 0.030). Poly-tobacco use involving all three products was associated with more than four times higher risk of cancer (aOR 4.54, 2.65–7.88; p < 0.001) [Table 3 & Supplementary Fig. S3]

Table 3.

Adjusted associations of bidi smoking and other tobacco-use patterns with cancer outcome using Firth penalized logistic regression.

Cancer outcome (Case vs. Control)	Cases/Control	aOR	95% CI	p-value
No habit	1429/1702	Ref (1.0)	–	–
Bidi only	82/30	3.11	1.86–5.31	<0.001a
Smokeless tobacco only	329/260	1.55	1.24–1.95	<0.001a
Cigarettes only	30/13	2.14	0.96–4.94	0.062
Bidi and smokeless tobacco	59/19	3.88	2.11–7.36	<0.001a
Bidi and cigarette	17/3	2.02	0.58–8.85	0.283
Cigarette and smokeless tobacco	20/10	2.64	1.10–6.58	0.030a
Poly-tobacco use (all three use)	71/29	4.54	2.65–7.88	<0.001a

aOR = adjusted odds ratio; CI = confidence interval. Cancer outcome (case/control) was modelled using multivariable Firth-penalised logistic regression. Models were adjusted for age, biological sex, education, occupation, household income, dietary pattern, body mass index, religion, hypertension, diabetes, and alcohol use. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

A clear exposure–response gradient was observed for both duration and daily frequency of bidi smoking. Compared with non-tobacco users (reference category), participants reporting <10 years of bidi use had nearly twice the odds of cancer (aOR 1.92, 95% CI 1.17–3.18; p = 0.010), whereas those with ≥10 years of use demonstrated almost seven times higher odds (aOR 6.75, 4.43–10.47; p < 0.001). A similar graded pattern was evident for smoking intensity. Individuals smoking <10 bidis per day had more than twice the odds of cancer (aOR 2.30, 1.47–3.61; p < 0.001), while those consuming ≥10 bidis daily exhibited over seven times higher odds (aOR 7.19, 4.57–11.51; p < 0.001) [Table 4]. Higher exposure categories demonstrated greater odds compared with lower exposure groups, indicating a marked escalation in risk with increasing duration and intensity of bidi use [Supplementary Fig. S4]. A statistically significant linear trend was observed for both duration and daily frequency of bidi use (p < 0.001). Sensitivity analyses using continuous measures of bidi exposure demonstrated consistent positive associations with cancer risk for both duration of use (aOR 1.04; 95% CI 1.03–1.06; p < 0.001) and daily frequency (aOR 1.08; 95% CI 1.05–1.11; p < 0.001) [Supplementary Table S6].

Table 4.

Dose-response associations of bidi smoking duration and daily frequency with cancer risk using Firth penalized logistic regression.

Outcome variables	Cases/Control	aOR	95% CI	p-value
Duration of bidi use
Non-tobacco users	1429/1702	Ref (1.0)	–	–
<10 years	60/42	1.92	1.17–3.18	0.010a
≥10 years	169/39	6.75	4.43–10.47	<0.001a
Frequency of bidi use per day
Non-tobacco users	1429/1702	Ref (1.0)	–	–
<10 per day	144/35	2.30	1.47–3.61	<0.001a
≥10 per day	85/46	7.19	4.57–11.51	<0.001a

aOR = adjusted odds ratio; CI = confidence interval. Duration and daily frequency of bidi smoking were modelled using multivariable Firth-penalised logistic regression. Models were adjusted for age, biological sex, education, occupation, household income, dietary pattern, body mass index, religion, hypertension, diabetes, and alcohol use. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

Metastasis was present in 8.0% of non-users, compared with 9.8% of exclusive bidi users and 11.9% of exclusive smokeless tobacco users. Higher proportions were observed among cigarette-only users (20.0%) and bidi–cigarette dual users (17.6%). The highest proportion of metastasis was noted among cigarette and smokeless tobacco dual users (35.0%). In contrast, poly-tobacco users demonstrated a metastasis prevalence of 7.0%. The overall association between tobacco-use pattern and metastasis was statistically significant (Pearson’s χ² = 149.13, p < 0.001), with a small effect size (Cramér’s V = 0.135). A significant linear trend was also observed (Cochran–Armitage χ² = 23.42, p < 0.001), indicating a graded increase in metastatic risk across hierarchical exposure categories [Table 5].

Table 5.

Gradient analysis of metastasis risk across hierarchical tobacco exposure categories.

Tobacco-use pattern	Metastasis absent n (%)	Metastasis PRESENT n (%)
No habit	1314 (92.0)	115 (8.0)
Bidi only	74 (90.2)	8 (9.8)
Smokeless tobacco only	290 (88.1)	39 (11.9)
Cigarettes only	24 (80.0)	6 (20.0)
Bidi and smokeless tobacco	53 (89.8)	6 (10.2)
Bidi and cigarette	14 (82.4)	3 (17.6)
Cigarette and smokeless tobacco	13 (65.0)	7 (35.0)
Poly-tobacco use (all three use)	66 (93.0)	5 (7.0)

Test of independence and dose–response evaluation

Pearson’s χ2 (df = 14) = 149.13, p < 0.001a Cramér’s V = 0.135 Cochran–Armitage trend χ2 (df = 1) = 23.42, p < 0.001a Fisher’s exact test (Monte Carlo simulated): p < 0.001a

df = degrees of freedom. Values represent row percentages within tobacco-use categories among participants with definitive metastasis status. Pearson’s chi-square test assessed independence between tobacco-use pattern and metastatic status. Cramér’s V quantified effect size. Cochran–Armitage trend test evaluated dose–response association across hierarchical tobacco-use categories. Fisher’s exact test with Monte Carlo simulation was additionally performed to confirm robustness of the association in the presence of sparse cell counts. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

Among non-users, dual-modality treatment was most frequent (38.8%), followed by single-modality treatment (30.0%). In contrast, palliative or supportive treatment was relatively more common among exclusive bidi users (37.8%) and bidi–cigarette dual users (35.3%). Multimodal treatment was more frequently observed among bidi–smokeless tobacco dual users (20.3%) and cigarette–smokeless tobacco users (25.0%) compared with non-users (10.9%). The association between tobacco use and treatment modalities was statistically significant (Pearson’s χ² = 215.83, p < 0.001; Cramér’s V = 0.115), with confirmation from the Kruskal–Wallis test demonstrating differences in ordinal treatment modalities (χ² = 125.49, p < 0.001) [Table 6].

Table 6.

Variability in oncologic treatment modalities across hierarchical tobacco exposure categories.

Tobacco-Use Pattern	Single treatment n (%)	Dual treatment n (%)	Multimodal treatment n (%)	Supportive/palliative care n (%)
No habit	429 (30.0)	555 (38.8)	156 (10.9)	289 (20.2)
Bidi only	26 (31.7)	18 (22.0)	7 (8.5)	31 (37.8)
Smokeless tobacco only	125 (38.0)	102 (31.0)	63 (19.1)	39 (11.9)
Cigarettes only	8 (26.7)	7 (23.3)	5 (16.7)	10 (33.3)
Bidi and smokeless tobacco	10 (16.9)	19 (32.2)	12 (20.3)	18 (30.5)
Bidi and cigarette	5 (29.4)	4 (23.5)	2 (11.8)	6 (35.3)
Cigarette and smokeless tobacco	5 (25.0)	4 (20.0)	5 (25.0)	6 (30.0)
Poly-tobacco use (all three use)	24 (33.8)	23 (32.4)	6 (8.5)	18 (25.4)

Comparative analysis of treatment modalities

Pearson’s χ2 (df = 28) = 215.83, p < 0.001a Cramér’s V = 0.115 Kruskal–Wallis χ2 (df = 7) = 125.49, p < 0.001a

df = degrees of freedom. Values represent row percentages within tobacco-use categories according to treatment modalities. Pearson’s chi-square test assessed independence between tobacco-use pattern and treatment modality. Cramér’s V quantified effect size. The Kruskal–Wallis test evaluated ordinal differences in treatment modalities across tobacco-use categories. Statistical significance was set at p ≤ 0.05.

^aIndicates statistical significance.

Discussion

The present multicentre case-control study examined differential cancer risk associated with various tobacco products, including bidi, in India. The results indicate that bidi smoking was associated with the highest adjusted odds of cancer among the tobacco products examined, consistent with prior Indian epidemiological studies.^4,37 Unlike factory-manufactured cigarettes, bidis are hand-rolled with unfiltered tendu leaves (Diospyros melanoxylon), resulting in incomplete combustion and higher concentrations of nicotine, tar, and carbon monoxide per unit of tobacco consumed.³⁸ A population-based study from Mumbai, India demonstrated that bidi smokers had a threefold higher risk of oral cancers compared to cigarette smokers.³ Similar results were reported by Gupta et al., who found that the relative risk of upper aerodigestive tract cancers was higher among bidi smokers than among cigarette smokers.³⁹ The current study’s adjusted estimate aligns closely with these observations and extends the evidence to multiple cancer types, suggesting that the toxic constituents and deeper inhalation patterns associated with bidi smoke may amplify systemic carcinogenic exposure.

Smokeless tobacco also showed a markedly elevated risk, which is consistent with global evidence linking tobacco use to multiple cancer types.^40,41 Data from the Global Burden of Disease 2019 study estimated that tobacco use accounted for 7.69 million deaths worldwide, with a substantial proportion attributable to cancer.¹ The carcinogenic mechanisms involve a complex interplay of nitrosamines, polycyclic aromatic hydrocarbons, and oxidative stress pathways leading to DNA adduct formation and genetic mutations.⁴² IARC monograph on smokeless tobacco has similarly confirmed its causal association with oral and pancreatic cancers.⁴³ The findings from the current study are consistent with the global evidence synthesised by Gupta et al., which demonstrated substantial variation in oral cancer risk across different smokeless tobacco products, with reported odds ratios ranging from 4.7 to 38.7 depending on product type. In comparison, the present study observed a comparatively lower magnitude of association for overall smokeless tobacco use (aOR 1.55), likely reflecting aggregation across diverse product types and multiple cancer sites.⁴⁴

A graded increase in cancer risk was observed in the present study with combined tobacco use. Participants who used both bidi and smokeless tobacco, as well as poly-tobacco users, demonstrated a significantly higher risk of developing cancer. These findings are consistent with a dose–response pattern, suggesting that cumulative carcinogen exposure from multiple tobacco forms may exert synergistic effects. Prior studies similarly reported higher odds of oral and respiratory diseases among dual tobacco users.^45,46 From a biological perspective, simultaneous exposure to smoke-derived carcinogens and nitrosamines from smokeless forms likely amplifies oxidative DNA damage, inhibits repair mechanisms, and accelerates oncogenic transformation.⁴⁶ These findings highlight the importance of addressing poly-tobacco use in both surveillance and cessation programmes, as single-product interventions may inadequately capture the total burden of exposure.

A clear dose–response relationship was observed between bidi smoking exposure and cancer risk. In the present study, increasing duration (≥10 years) and higher daily frequency (≥10 bidis/day) were associated with stepwise increases in adjusted cancer odds, indicating cumulative carcinogenic burden. This gradient is biologically plausible and consistent with large Indian cohort and case–control studies reported in the literature. The Mumbai cohort study demonstrated a monotonic increase in site-specific cancer incidence with escalating bidi consumption, with heavier smokers exhibiting higher risks than light users, particularly for cancers of the oral cavity, pharynx, and lung.³ Similar exposure–response trends were reported in the Karunagappally cohort (Kerala, India), where long-term bidi use showed progressively increasing hazard ratios for upper aerodigestive tract cancers.¹⁵ Bidi use generates higher per-unit exposure to tobacco smoke toxicants due to the absence of filtration and inefficient combustion, which necessitate deeper and more frequent inhalation, thereby progressively amplifying cumulative DNA damage with increasing intensity and duration of use.³⁹ The consistency of this dose gradient across epidemiological designs, populations, and cancer sites supports the presence of an exposure–response continuum, suggesting that regulatory and cessation strategies should account for both intensity and chronicity of bidi use rather than binary smoking status alone.

Tobacco users also exhibited increased risk of metastatic cancer. Tobacco users often present at advanced stages due to delayed symptom recognition and compromised immune surveillance.⁴⁷ Nicotine and other tobacco-derived alkaloids are known to promote angiogenesis, epithelial-mesenchymal transition, and tumour cell migration through activation of the nicotinic acetylcholine receptor pathway.⁴⁸ The higher metastatic risk among multiple-product users may thus reflect both biological aggressiveness and behavioural factors such as lower health-seeking behaviour among heavy users. Participants who used multiple forms of tobacco required multimodal treatment approaches more often than non-users. This could be attributed to a combination of factors including advanced disease stage, multi-organ involvement, and poorer physiological reserves that complicate treatment planning. A previous study demonstrated that continued tobacco use among cancer patients impairs treatment efficacy, increases the likelihood of recurrence, and exacerbates therapy-related toxicities.⁴⁹ In the Indian context, where treatment adherence is often influenced by socioeconomic constraints, tobacco use may indirectly contribute to poorer outcomes through delayed diagnosis and suboptimal response to therapy. The relatively higher proportion of dual-modality treatment observed among non-tobacco users is likely attributable to random variation and small subgroup distributions, rather than a true underlying association.

Globally, smoking prevalence has declined in many high-income countries due to sustained tobacco control policies, yet the burden of smokeless and poly-tobacco use remains concentrated in South Asia.⁵⁰ The observed adjusted odds ratios in this study are higher than those reported in cohorts from other developed countries, possibly reflecting differences in product composition, absence of filters, and higher concentrations of unburned organic material in bidis and locally prepared smokeless forms.^3,51 These findings carry important implications for national cancer control and tobacco regulation. India’s Cigarettes and Other Tobacco Products Act (COTPA) and the National Tobacco Control Programme have achieved measurable progress in curbing cigarette consumption, but their impact on bidis has been limited. Recent assessments using the Tobacco Control Scale revealed that India performs poorly in the enforcement of policies pertaining to non-cigarette tobacco products.⁵² This study’s demonstration of high cancer risk associated with bidis and should prompt regulatory recalibration to include standardised packaging, taxation parity, and targeted cessation interventions for rural and informal sector populations. Moreover, integrating routine tobacco-use screening including bidi into primary care setting could facilitate early cessation support and improve treatment outcomes.

While the present study represents a multicentre analysis on bidi and cancer in India, certain limitations merit consideration. The hospital-based design may introduce selection bias as controls were drawn from hospital visitors rather than the general population. The lack of individual matching may introduce residual confounding, although key covariates were adjusted for in multivariable analyses. Self-reported tobacco use may be subject to recall bias or underreporting, particularly among women due to social desirability concerns. The case–control design precludes direct temporal inference between exposure and outcome. Additionally, although key confounders such as BMI, diabetes, and hypertension were adjusted for, residual confounding from environmental pollutants or dietary micronutrient deficiencies cannot be fully excluded. The estimated odds ratios in the present study are not directly comparable to cohort-derived measures; however, comparisons were made due to limited evidence on bidi smoking in the Indian setting. Residual bias due to missing data cannot be fully excluded, as missingness may not be completely random despite the use of a missing category approach. Site-specific analyses for individual cancer types could not be performed due to heterogeneity and limited sample sizes within specific cancer categories, which may lead to unstable estimates. Despite these limitations, the large sample size, standardized data collection, and multi-regional representation lend substantial credibility to the findings.

Future research should focus on using biochemical validation such as cotinine assays, although their utility in case–control studies is limited for capturing long-term exposure. Investigating gene-environment interactions, particularly polymorphisms in metabolic enzymes, could elucidate inter-individual susceptibility patterns. In addition, implementation research evaluating the effectiveness of context-specific cessation strategies for bidi and smokeless users is essential.

In this hospital-based unmatched case-control study across seven Indian states, bidi smoking was associated with the highest adjusted odds of cancer among the tobacco products examined. A higher risk of cancer was observed among individuals engaging in dual or poly-tobacco use, particularly combinations involving bidis. The findings further indicate that tobacco exposure, especially bidi smoking is strongly associated with adverse clinical characteristics. Poly-tobacco users presented more often with metastatic disease at diagnosis and required more intensive multimodal cancer treatment. The presence of clear dose response relationships with both duration and daily frequency of bidi use highlights the cumulative harm associated with sustained exposure. Overall, the study identifies bidi smoking as a disproportionately high-risk form of tobacco use in India, with implications extending beyond cancer occurrence to disease severity and treatment burden. These findings highlight the need to explicitly address bidis and poly-tobacco use when designing cancer prevention, early detection, and tobacco cessation strategies in the Indian context.

Contributors

UV: Conceptualization, Methodology, Formal analysis, Supervision, Validation, Data curation, Project administration, Writing – original draft, Writing – review & editing; OPB: Conceptualization, Methodology, Supervision, Funding, Validation, Data curation, Writing – original draft, Writing – review & editing; VNS: Methodology, Visualization, Formal analysis, Data curation, Writing – original draft, Writing – review & editing; PA: Methodology, Validation, Data curation, Writing – original draft, Writing – review & editing; ACB: Methodology, Formal analysis, Visualization, Writing – original draft, Writing – review & editing; AM: Methodology, Formal analysis, Visualization, Writing – original draft, Writing – review & editing; FA: Methodology, Validation, Data curation, Writing – review & editing; CS: Methodology, Validation, Data curation, Writing – review & editing; KJ: Methodology, Validation, Data curation, Writing – review & editing; BNN: Methodology, Validation, Data curation, Writing – review & editing; HB: Methodology, Validation, Data curation, Writing – review & editing; SS: Methodology, Formal analysis, Visualization, Writing – original draft, Writing – review & editing; CM: Methodology, Validation, Data curation, Writing – review & editing; MJ: Methodology, Validation, Data curation, Writing – review & editing; BB: Methodology, Validation, Data curation, Writing – review & editing; KG: Methodology, Validation, Data curation, Writing – review & editing; HSJ: Methodology, Validation, Data curation, Writing – review & editing; VD: Methodology, Validation, Data curation, Writing – review & editing. All authors reviewed the manuscript and approved the final version. All authors had full access to all the data in the study and accept responsibility for the decision to submit the manuscript for publication.

Data sharing statement

The datasets analysed during the current study are not publicly available due to ethical and institutional restrictions but are available from the corresponding author upon reasonable request. Requests for access to the data should include a clear description of the research purpose and proposed analysis plan. Data will be shared in accordance with institutional guidelines, ethical approvals, and applicable data protection regulations.

Editor’s note

The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations.

Statement on AI use

The authors declare that during preparation of this manuscript ChatGPT was used to refine language and improve clarity of the manuscript. After using this tool/service, the authors reviewed and edited the content and take full responsibility for accuracy and integrity of the work.

Declaration of interests

All authors declare that they have no competing interests.