Association of Sulfur Microbial Diet and the Risk of Lung Cancer in Women Authors: Gaofeng Liang, Hongyan Yu, Chengbin Lin, Jinxian He, Hui Tian Published: 2026 ABSTRACT PurposeLung cancer is the leading cause of cancer death worldwide, and its burden among women—including women who have never smoked—continues to rise. The sulfur microbial diet (SMD), a dietary pattern linked to the abundance of sulfur-metabolizing gut bacteria that generate genotoxic hydrogen sulfide, has been associated with colorectal neoplasia, but whether it is associated with lung cancer risk in women is unknown.Patients and MethodsWe conducted a prospective cohort study of 114,981 women with valid 24-hour dietary assessments. An SMD score was derived from eight dietary components, and participants were classified into quartiles (Q1–Q4). Incident lung cancer was ascertained through cancer-registry linkage. Cox proportional-hazards models with sequential adjustment for sociodemographic, lifestyle, clinical and dietary factors estimated hazard ratios (HRs) and 95% confidence intervals (CIs). Restricted cubic spline, subgroup, sensitivity, substitution and mediation analyses were also performed.ResultsDuring follow-up, 802 incident lung cancer cases were documented. In the fully adjusted model, women in the highest SMD quartile had a higher risk of lung cancer than those in the lowest quartile (HR 1.25, 95% CI 1.02–1.54; P for trend = 0.036), and each 1-standard-deviation increase was associated with an 8% higher risk (HR 1.08, 95% CI 1.00–1.16). The dose–response association was positive and approximately linear, and the findings were robust across subgroup and sensitivity analyses. Substituting vegetables for processed meat was associated with a non-significantly lower risk (HR 0.93, 95% CI 0.86–1.01), and metabolic and inflammatory biomarkers showed minimal mediation.ConclusionGreater adherence to a more sulfur-microbial-promoting diet was associated with a modestly higher risk of incident lung cancer in women. These findings extend the relevance of the SMD beyond colorectal neoplasia and support dietary patterns favouring vegetables and legumes over processed meat for lung cancer prevention. Introduction Lung cancer remains the most commonly diagnosed cancer and the leading cause of cancer death worldwide, accounting for an estimated 2.5 million new cases and 1.8 million deaths in 2022.1 The great majority of cases are non-small cell lung cancer, and despite advances in early detection and systemic therapy, the prognosis remains poor for most patients diagnosed at an advanced stage.2 Although lung cancer has historically been regarded as a disease of men, it now imposes a substantial and, in many populations, growing burden among women, in whom it ranks among the leading causes of cancer-related death.1,3 Women also differ from men in the histological distribution, molecular characteristics and clinical course of lung cancer.3,4 Cigarette smoking is the dominant cause of lung cancer in both sexes,5 yet a considerable and increasing proportion of cases in women—variously estimated at 15% to 25%—occurs in lifelong never-smokers.6,7 Together, these observations underscore the need to identify modifiable, non-tobacco risk factors that could inform prevention in women. Diet is a widely shared and modifiable exposure that has long been implicated in lung carcinogenesis. Prospective studies have consistently linked higher intakes of fruits, vegetables and dietary fibre to lower lung cancer risk,8,9 whereas higher consumption of red and processed meat has been associated with increased risk.9,10 The International Agency for Research on Cancer has classified processed meat as carcinogenic and red meat as probably carcinogenic to humans,11 a judgement attributed in part to N-nitroso compounds, heterocyclic amines, polycyclic aromatic hydrocarbons and other contaminants generated during processing and cooking.12 Cruciferous vegetables and their isothiocyanate constituents have likewise been associated with lower risk,13 whereas randomised trials of β-carotene supplementation unexpectedly increased lung cancer incidence among smokers,14 illustrating that the relationship between diet and lung cancer is complex and cannot be reduced to single nutrients. Because foods are consumed in combination, dietary pattern analyses that capture the joint effects of multiple foods may better reflect the biological consequences of habitual diet.9 An emerging body of evidence implicates the gut microbiome in carcinogenesis and in the host response to cancer.15,16 Through the so-called gut–lung axis, the intestinal microbiota modulate systemic immunity and inflammation and can thereby influence the lung, a distant mucosal site.17 One mechanistically defined link between diet and the microbiome involves sulfur-metabolizing (sulfate-reducing) bacteria, which reduce dietary sulfur to hydrogen sulfide, a gas that is genotoxic to epithelial cells and promotes mucosal inflammation.18–20 Because the abundance of these bacteria is shaped by diet, Nguyen et al derived the sulfur microbial diet (SMD)—an a priori dietary pattern correlated with the faecal abundance of sulfur-metabolizing bacteria and characterised by higher intakes of processed meat and low-calorie beverages and lower intakes of vegetables and legumes.21 Greater adherence to the SMD has since been associated with an increased risk of colorectal cancer,22 early-onset colorectal cancer precursors,23 colorectal cancer of specific molecular subtypes24 and colorectal adenoma in older adults,25 as well as with higher all-cause and cancer mortality.26 To date, however, the SMD has been examined almost exclusively in relation to colorectal outcomes, and whether a sulfur-microbial-promoting diet is associated with the risk of extra-intestinal malignancies—and lung cancer in particular—remains unknown. Such an association is biologically plausible: gut microbial dysbiosis and the metabolites it generates can influence pulmonary immune surveillance and inflammation through the gut–lung axis,17 systemic inflammatory mediators have been linked to lung tumour progression,27 and hydrogen-sulfide-related signalling has been implicated in tumour growth and metastasis in experimental models.28 Women merit dedicated investigation, given the distinct epidemiology and hormonal context of lung cancer in women4,29 and known sex differences in the gut microbiome. Nevertheless, no prospective study has evaluated the relationship between the SMD and lung cancer risk in women. We therefore used prospective cohort to investigate the association of the SMD with incident lung cancer among women. Specifically, we (i) estimated the prospective association between the SMD score and incident lung cancer using Cox proportional-hazards models; (ii) characterised the dose–response relationship using restricted cubic splines; (iii) assessed the robustness of the findings through subgroup, sensitivity and food-substitution analyses; and (iv) explored whether circulating metabolic and inflammatory biomarkers mediate the association. Methods Study Design and Population This prospective cohort study was based on a large population-based cohort that enrolled approximately 500,000 adults aged 40–69 years across the United Kingdom between 2006 and 2010.30 At recruitment, participants provided written informed consent and completed extensive baseline assessments; this study has ethical approval from the North West Multi-centre Research Ethics Committee, and the present analysis was conducted under approved access (78619). Participants with at least one valid 24-hour dietary assessment were eligible for the present study. The analysis was restricted to women; male participants were excluded. We further excluded women with prevalent lung cancer at baseline and those with missing follow-up information or a follow-up time of zero years or less. After these exclusions, 114,981 women were included in the final analysis, among whom 802 incident lung cancer cases were identified during follow-up. The participant selection process is summarised in Figure 1. Figure 1Flow diagram of participant selection.The flowchart illustrates the participant selection process for a study. Initially, 501,936 participants were recruited between 2006 and 2010. From these, 291,197 participants with no reliable 24-hour diet assessment data were excluded, leaving 210,739 participants. Next, 94,659 male participants were excluded, resulting in 116,080 participants. Then, 42 participants who had lung cancer before recruitment were excluded, reducing the number to 116,038 participants. Finally, participants with unavailable dietary data required for SMD calculation, totaling 1,057, were excluded. This led to a final inclusion of 114,981 participants in the study.A flowchart of participant selection in a study, detailing exclusions and final inclusion count. Flow diagram of participant selection. Assessment of the Sulfur Microbial Diet Score Dietary intake was assessed using the Oxford WebQ, a web-based 24-hour dietary recall instrument administered on up to five occasions.31 For each participant, the average intake of each food item across all completed assessments was calculated to better reflect habitual intake. The SMD score was derived from the dietary pattern originally identified by Nguyen et al as correlating with the faecal abundance of sulfur-metabolizing gut bacteria.21 Adapted to the foods captured by the Oxford WebQ, the score incorporated eight dietary components—processed meat, liquor, low-calorie drinks, beer, fruit juice, legumes, vegetables and sweets/desserts. Weighted coefficients were applied to each component, with a higher score indicating greater adherence to a more sulfur-microbial-promoting dietary pattern. Participants were categorised into quartiles (Q1–Q4) according to the distribution of the SMD score, and a standardised SMD z-score was additionally generated for continuous analyses. Outcome Ascertainment The primary outcome was incident lung cancer, identified through linkage with national cancer registries using the International Classification of Diseases, Tenth Revision (ICD-10) codes C33–C34 and the Ninth Revision (ICD-9) code 162. The earliest recorded diagnosis date was taken as the event date. Follow-up time was calculated from the date of baseline assessment until the date of lung cancer diagnosis, death, loss to follow-up, or 1 January 2024, whichever occurred first. Participants were recruited from 2006 to 2010 at baseline, corresponding to a maximum follow-up duration of around 18 years. Covariates Potential confounders were selected a priori on the basis of established or plausible associations with both diet and lung cancer. These comprised age, ethnicity, body mass index (BMI), Townsend deprivation index, educational attainment, household income, employment status, smoking status, alcohol intake, physical activity, diabetes, cardiovascular disease (CVD), hypertension, use of blood-pressure-lowering medication, use of lipid-lowering medication, insulin use and total energy intake. Missing covariate data were addressed using multiple imputation by chained equations, generating five imputed datasets.32 Multiple imputation by chained equations was performed with sufficient iterations to reach convergence (chain parameter fluctuation <0.1%); five imputed datasets were pooled via Rubin’s rules to generate combined HR and 95% CI. Notably, validated variables for passive smoking, ambient air pollution and occupational toxic exposure were not collected at baseline assessment and thus could not be included as confounding covariates in regression models. Detailed category definitions for all categorical confounding variables are described as follows: Ethnicity: White, Non-white; Educational attainment: College/University, A/O levels, Other qualifications, None; Household income: Less than 18,000, 18,000 to 30,999, 31,000 to 51,999, 52,000 to 100,000, Greater than 100,000; Employed: binary variable (yes/no); Smoking status: Never, Previous, Current; Physical activity: Low, Moderate, High; Binary disease indicators (Diabetes, Cardiovascular disease, Hypertension): yes/no; Insulin use: binary variable (yes/no). Statistical Analysis Baseline characteristics were summarised across quartiles of the sulfur microbial diet score and according to incident lung cancer status. Continuous variables are presented as mean ± standard deviation (SD), and categorical variables as numbers and percentages. For the comparison between participants who developed incident lung cancer during follow-up and those who remained free of lung cancer, absolute standardized differences were calculated to describe the magnitude of between-group differences. The association between the SMD score and incident lung cancer was evaluated using Cox proportional-hazards regression. Three sequential models were constructed: Model 1 was adjusted for age and ethnicity; Model 2 was additionally adjusted for socioeconomic and lifestyle factors (BMI, Townsend deprivation index, education, income, employment, smoking status, alcohol intake, physical activity, diabetes, CVD and hypertension); and Model 3 was further adjusted for blood-pressure-lowering medication, lipid-lowering medication, insulin use and total energy intake. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated with the lowest quartile (Q1) as the reference and were pooled across the imputed datasets using Rubin’s rules. Tests for linear trend were performed by assigning the median SMD value within each quartile and modelling it as a continuous variable, and the SMD score was also modelled continuously (per 1-SD increment). The proportional hazards assumption was verified via Schoenfeld residual test, no significant violation was observed (P>0.05). Restricted cubic spline models were used to examine the dose–response relationship between the continuous SMD score and lung cancer risk, with formal tests for non-linearity.33 Restricted cubic spline models were fitted with 4 knots placed at the 5th, 35th, 65th and 95th percentiles of continuous SMD score. Pre-specified subgroup analyses were conducted according to age, ethnicity, alcohol intake and smoking status, with multiplicative interaction assessed by likelihood-ratio tests. The robustness of the findings was examined in sensitivity analyses that excluded lung cancer events occurring within the first two years of follow-up, to address potential reverse causation, and that excluded women with diabetes or CVD at baseline. Substitution analyses estimated the theoretical effect of replacing one dietary component with an equivalent amount of another while holding total intake constant.34 Finally, approximate mediation analyses were performed to explore the potential mediating roles of circulating metabolic and inflammatory biomarkers, including BMI, glucose, high-density lipoprotein (HDL) cholesterol, total cholesterol, triglycerides, systolic and diastolic blood pressure, and C-reactive protein. Approximate mediation analysis was conducted separately for each single circulating biomarker; direct effect, indirect effect and proportion mediated were calculated by sequential regression adjustment, mutual correction between multiple mediators was not performed. The mediation testing framework adopted herein followed the sequential regression-based approximate mediation approach. First, the total effect of SMD score on lung cancer risk was estimated without adjusting for any circulating biomarker. Second, the direct effect was calculated after additionally adjusting for one single biomarker mediator. The indirect mediating effect was derived by comparing the total effect and direct effect. The percentage mediated proportion was quantified to reflect the share of the association explained by each biomarker. All mediation-related HRs and mediated proportions were pooled across multiply imputed datasets in accordance with Rubin’s rules. Each circulating biomarker was evaluated in an independent single-mediator regression model rather than one unified model containing all mediators simultaneously. Separate regression models were constructed for each biomarker owing to variable-specific missing values, which could lead to minor differences in effective sample size across different mediation models. All statistical tests were two-sided, and P < 0.05 was considered statistically significant. All statistical analyses were conducted using R software version 4.3.3. Cox proportional-hazards models were implemented with the survival package; restricted cubic spline analysis was performed via the rms package; multiple imputation was completed using the mice package. Results Baseline Characteristics A total of 114,981 women were included in the analysis. Compared with women in the lowest quartile of the SMD score, those in the highest quartile were on average younger, had a higher BMI, and were more likely to be current smokers and physically inactive and to have diabetes, CVD and hypertension; they also had lower educational attainment, lower household income and a lower mean total energy intake. Baseline characteristics across SMD quartiles are presented in Table 1. During follow-up, 802 women developed incident lung cancer, whereas 114,179 remained free of lung cancer. Compared with women who remained free of lung cancer, those who developed lung cancer were older (60.24 vs 55.61 years), had a higher Townsend deprivation index (−0.90 vs −1.56), were less likely to have a college/university education (29.6% vs 41.2%) or to be currently employed (43.4% vs 59.9%), and were more likely to be previous or current smokers (76.2% vs 38.9%). They also had higher prevalences of diabetes (4.6% vs 2.9%), CVD (40.4% vs 22.0%), and hypertension (28.7% vs 20.9%), and were more likely to use blood-pressure-lowering (22.1% vs 13.9%) and lipid-lowering medications (20.3% vs 11.4%) (all P ≤ 0.001). In contrast, ethnicity, BMI, insulin use, total energy intake, the SMD z-score, and SMD quartile distribution did not differ significantly between the two groups (all P > 0.05). Detailed baseline characteristics according to incident lung cancer status are presented in Table 2.Table 1Baseline Characteristics of Participants According to Quartiles of the Sulfur Microbial Diet ScoreCharacteristicTotal(N = 114,981)Q1(n = 28,810)Q2(n = 28,705)Q3(n = 28,740)Q4(n = 28,726)Age, years55.65 ± 7.8456.06 ± 7.6856.08 ± 7.7455.70 ± 7.8854.78 ± 7.99BMI, kg/m226.55 ± 5.0326.05 ± 4.8426.26 ± 4.8626.51 ± 4.9127.37 ± 5.38Townsend deprivation index−1.56 ± 2.85−1.57 ± 2.82−1.66 ± 2.80−1.60 ± 2.83−1.40 ± 2.93Total energy intake, kJ/day8,257.6 ± 2,423.78,961.3 ± 2,431.28,415.3 ± 2,280.48,056.9 ± 2,359.67,594.8 ± 2,411.9Ethnicity, n (%) White110,027 (95.69)27,736 (96.27)27,670 (96.39)27,479 (95.61)27,142 (94.49) Non-white4,954 (4.31)1,074 (3.73)1,035 (3.61)1,261 (4.39)1,584 (5.51)Educational attainment, n (%) College/University47,617 (41.41)13,672 (47.46)12,499 (43.54)11,504 (40.03)9,942 (34.61) A/O levels42,347 (36.83)9,887 (34.32)10,450 (36.41)10,835 (37.70)11,175 (38.90) Other qualifications15,362 (13.36)3,470 (12.05)3,527 (12.29)3,869 (13.46)4,496 (15.65) None9,655 (8.40)1,780 (6.18)2,229 (7.77)2,532 (8.81)3,113 (10.84)Household income, n (%) Less than 18,00021,151 (18.40)5,166 (17.93)5,070 (17.66)5,284 (18.39)5,631 (19.60) 18,000 to 30,99929,739 (25.86)7,552 (26.21)7,408 (25.81)7,426 (25.84)7,353 (25.60) 31,000 to 51,99931,452 (27.35)7,781 (27.01)7,988 (27.83)7,952 (27.67)7,730 (26.91) 52,000 to 100,00025,452 (22.14)6,483 (22.50)6,371 (22.19)6,294 (21.90)6,303 (21.94) Greater than 100,0007,187 (6.25)1,827 (6.34)1,868 (6.51)1,783 (6.20)1,709 (5.95) Employed, n (%)68,767 (59.81)16,865 (58.54)16,755 (58.37)17,095 (59.48)18,052 (62.84)Smoking status, n (%) Never69,894 (60.79)17,635 (61.21)17,764 (61.88)17,760 (61.80)16,735 (58.26) Previous37,435 (32.56)9,669 (33.56)9,249 (32.22)9,063 (31.54)9,454 (32.91) Current7,652 (6.65)1,506 (5.23)1,692 (5.89)1,916 (6.67)2,538 (8.83)Physical activity, n (%) Low40,807 (35.49)9,278 (32.21)10,108 (35.21)10,395 (36.17)11,025 (38.38) Moderate46,859 (40.75)11,857 (41.16)11,812 (41.15)11,712 (40.75)11,477 (39.95) High27,315 (23.76)7,674 (26.64)6,784 (23.63)6,632 (23.08)6,224 (21.67) Diabetes, n (%)3,373 (2.93)689 (2.39)778 (2.71)824 (2.87)1,082 (3.77) Cardiovascular disease, n (%)25,409 (22.10)6,029 (20.93)6,199 (21.59)6,241 (21.71)6,941 (24.16) Hypertension, n (%)24,094 (20.95)5,753 (19.97)5,872 (20.46)5,906 (20.55)6,564 (22.85) Insulin use, n (%)726 (0.63)139 (0.48)154 (0.54)181 (0.63)252 (0.88)Notes: Data are presented as mean ± standard deviation for continuous variables and n (%) for categorical variables; values were pooled across five multiply imputed datasets. Q1–Q4, quartiles of the sulfur microbial diet score.Abbreviation: BMI, body mass index. Table 2Baseline Characteristics According to Incident Lung Cancer StatusCharacteristicOverallN=114,981Non-CasesN=114,179Lung CancerN=802P valueAge, years55.65 ± 7.8455.61 ± 7.8460.24 ± 6.23<0.001Ethnicity0.060 White110,176 (95.8)109,402 (95.8)774 (96.5) Non-white4,482 (3.9)4,459 (3.9)23 (2.9) Missing/unknown323 (0.3)318 (0.3)5 (0.6) BMI, kg/m^226.55 ± 5.0326.55 ± 5.0326.84 ± 4.910.101 Townsend deprivation index−1.56 ± 2.85−1.56 ± 2.85−0.90 ± 2.83<0.001Educational attainment<0.001 College/university47,321 (41.2)47,084 (41.2)237 (29.6) A/O levels or equivalent42,125 (36.6)41,840 (36.6)285 (35.5) Other qualifications15,318 (13.3)15,203 (13.3)115 (14.3) None9,657 (8.4)9,500 (8.3)157 (19.6) Missing/unknown560 (0.5)552 (0.5)8 (1.0)Household income, GBP/year<0.001 Less than 18,00017,317 (15.1)17,094 (15.0)223 (27.8) 18,000 to 30,99925,582 (22.2)25,373 (22.2)209 (26.1) 31,000 to 51,99927,971 (24.3)27,831 (24.4)140 (17.5) 52,000 to 100,00022,757 (19.8)22,671 (19.9)86 (10.7) Greater than 100,0006,543 (5.7)6,524 (5.7)19 (2.4) Missing/unknown14,811 (12.9)14,686 (12.9)125 (15.6)Currently employed<0.001 No46,214 (40.2)45,760 (40.1)454 (56.6) Yes68,767 (59.8)68,419 (59.9)348 (43.4)Smoking status<0.001 Never69,646 (60.6)69,457 (60.8)189 (23.6) Previous37,334 (32.5)36,959 (32.4)375 (46.8) Current7,708 (6.7)7,472 (6.5)236 (29.4) Missing/unknown293 (0.3)291 (0.3)2 (0.2)Alcohol intake<0.001 Never4,848 (4.2)4,821 (4.2)27 (3.4) Previous3,592 (3.1)3,541 (3.1)51 (6.4) Current106,425 (92.6)105,703 (92.6)722 (90.0) Missing/unknown116 (0.1)114 (0.1)2 (0.2)Physical activity<0.001 Low15,971 (13.9)15,845 (13.9)126 (15.7) Moderate40,176 (34.9)39,940 (35.0)236 (29.4) High35,223 (30.6)35,005 (30.7)218 (27.2) Missing/unknown23,611 (20.5)23,389 (20.5)222 (27.7)Diabetes0.001 No111,421 (96.9)110,660 (96.9)761 (94.9) Yes3,358 (2.9)3,321 (2.9)37 (4.6) Missing/unknown202 (0.2)198 (0.2)4 (0.5)Cardiovascular disease<0.001 No89,572 (77.9)89,094 (78.0)478 (59.6) Yes25,409 (22.1)25,085 (22.0)324 (40.4)Hypertension<0.001 No90,887 (79.0)90,315 (79.1)572 (71.3) Yes24,094 (21.0)23,864 (20.9)230 (28.7)Blood-pressure-lowering medication<0.001 No98,884 (86.0)98,259 (86.1)625 (77.9) Yes16,097 (14.0)15,920 (13.9)177 (22.1)Lipid-lowering medication<0.001 No101,758 (88.5)101,119 (88.6)639 (79.7) Yes13,223 (11.5)13,060 (11.4)163 (20.3)Insulin use0.189 No114,255 (99.4)113,461 (99.4)794 (99.0) Yes726 (0.6)718 (0.6)8 (1.0) Total energy intake, kJ/day8,022.71 ± 2,209.318,022.68 ± 2,207.948,028.12 ± 2,398.430.949 Sulfur microbial diet z-score0.00 ± 1.00−0.00 ± 1.000.04 ± 1.060.273Sulfur microbial diet quartile0.275 Q128,810 (25.1)28,613 (25.1)197 (24.6) Q228,705 (25.0)28,520 (25.0)185 (23.1) Q328,740 (25.0)28,543 (25.0)197 (24.6) Q428,726 (25.0)28,503 (25.0)223 (27.8)Notes: Bold P values indicate statistically significant differences (P < 0.05) between participants with incident lung cancer and participants without lung cancer. Table 1Baseline Characteristics of Participants According to Quartiles of the Sulfur Microbial Diet ScoreCharacteristicTotal(N = 114,981)Q1(n = 28,810)Q2(n = 28,705)Q3(n = 28,740)Q4(n = 28,726)Age, years55.65 ± 7.8456.06 ± 7.6856.08 ± 7.7455.70 ± 7.8854.78 ± 7.99BMI, kg/m226.55 ± 5.0326.05 ± 4.8426.26 ± 4.8626.51 ± 4.9127.37 ± 5.38Townsend deprivation index−1.56 ± 2.85−1.57 ± 2.82−1.66 ± 2.80−1.60 ± 2.83−1.40 ± 2.93Total energy intake, kJ/day8,257.6 ± 2,423.78,961.3 ± 2,431.28,415.3 ± 2,280.48,056.9 ± 2,359.67,594.8 ± 2,411.9Ethnicity, n (%) White110,027 (95.69)27,736 (96.27)27,670 (96.39)27,479 (95.61)27,142 (94.49) Non-white4,954 (4.31)1,074 (3.73)1,035 (3.61)1,261 (4.39)1,584 (5.51)Educational attainment, n (%) College/University47,617 (41.41)13,672 (47.46)12,499 (43.54)11,504 (40.03)9,942 (34.61) A/O levels42,347 (36.83)9,887 (34.32)10,450 (36.41)10,835 (37.70)11,175 (38.90) Other qualifications15,362 (13.36)3,470 (12.05)3,527 (12.29)3,869 (13.46)4,496 (15.65) None9,655 (8.40)1,780 (6.18)2,229 (7.77)2,532 (8.81)3,113 (10.84)Household income, n (%) Less than 18,00021,151 (18.40)5,166 (17.93)5,070 (17.66)5,284 (18.39)5,631 (19.60) 18,000 to 30,99929,739 (25.86)7,552 (26.21)7,408 (25.81)7,426 (25.84)7,353 (25.60) 31,000 to 51,99931,452 (27.35)7,781 (27.01)7,988 (27.83)7,952 (27.67)7,730 (26.91) 52,000 to 100,00025,452 (22.14)6,483 (22.50)6,371 (22.19)6,294 (21.90)6,303 (21.94) Greater than 100,0007,187 (6.25)1,827 (6.34)1,868 (6.51)1,783 (6.20)1,709 (5.95) Employed, n (%)68,767 (59.81)16,865 (58.54)16,755 (58.37)17,095 (59.48)18,052 (62.84)Smoking status, n (%) Never69,894 (60.79)17,635 (61.21)17,764 (61.88)17,760 (61.80)16,735 (58.26) Previous37,435 (32.56)9,669 (33.56)9,249 (32.22)9,063 (31.54)9,454 (32.91) Current7,652 (6.65)1,506 (5.23)1,692 (5.89)1,916 (6.67)2,538 (8.83)Physical activity, n (%) Low40,807 (35.49)9,278 (32.21)10,108 (35.21)10,395 (36.17)11,025 (38.38) Moderate46,859 (40.75)11,857 (41.16)11,812 (41.15)11,712 (40.75)11,477 (39.95) High27,315 (23.76)7,674 (26.64)6,784 (23.63)6,632 (23.08)6,224 (21.67) Diabetes, n (%)3,373 (2.93)689 (2.39)778 (2.71)824 (2.87)1,082 (3.77) Cardiovascular disease, n (%)25,409 (22.10)6,029 (20.93)6,199 (21.59)6,241 (21.71)6,941 (24.16) Hypertension, n (%)24,094 (20.95)5,753 (19.97)5,872 (20.46)5,906 (20.55)6,564 (22.85) Insulin use, n (%)726 (0.63)139 (0.48)154 (0.54)181 (0.63)252 (0.88)Notes: Data are presented as mean ± standard deviation for continuous variables and n (%) for categorical variables; values were pooled across five multiply imputed datasets. Q1–Q4, quartiles of the sulfur microbial diet score.Abbreviation: BMI, body mass index. Baseline Characteristics of Participants According to Quartiles of the Sulfur Microbial Diet Score Notes: Data are presented as mean ± standard deviation for continuous variables and n (%) for categorical variables; values were pooled across five multiply imputed datasets. Q1–Q4, quartiles of the sulfur microbial diet score. Abbreviation: BMI, body mass index. Table 2Baseline Characteristics According to Incident Lung Cancer StatusCharacteristicOverallN=114,981Non-CasesN=114,179Lung CancerN=802P valueAge, years55.65 ± 7.8455.61 ± 7.8460.24 ± 6.23<0.001Ethnicity0.060 White110,176 (95.8)109,402 (95.8)774 (96.5) Non-white4,482 (3.9)4,459 (3.9)23 (2.9) Missing/unknown323 (0.3)318 (0.3)5 (0.6) BMI, kg/m^226.55 ± 5.0326.55 ± 5.0326.84 ± 4.910.101 Townsend deprivation index−1.56 ± 2.85−1.56 ± 2.85−0.90 ± 2.83<0.001Educational attainment<0.001 College/university47,321 (41.2)47,084 (41.2)237 (29.6) A/O levels or equivalent42,125 (36.6)41,840 (36.6)285 (35.5) Other qualifications15,318 (13.3)15,203 (13.3)115 (14.3) None9,657 (8.4)9,500 (8.3)157 (19.6) Missing/unknown560 (0.5)552 (0.5)8 (1.0)Household income, GBP/year<0.001 Less than 18,00017,317 (15.1)17,094 (15.0)223 (27.8) 18,000 to 30,99925,582 (22.2)25,373 (22.2)209 (26.1) 31,000 to 51,99927,971 (24.3)27,831 (24.4)140 (17.5) 52,000 to 100,00022,757 (19.8)22,671 (19.9)86 (10.7) Greater than 100,0006,543 (5.7)6,524 (5.7)19 (2.4) Missing/unknown14,811 (12.9)14,686 (12.9)125 (15.6)Currently employed<0.001 No46,214 (40.2)45,760 (40.1)454 (56.6) Yes68,767 (59.8)68,419 (59.9)348 (43.4)Smoking status<0.001 Never69,646 (60.6)69,457 (60.8)189 (23.6) Previous37,334 (32.5)36,959 (32.4)375 (46.8) Current7,708 (6.7)7,472 (6.5)236 (29.4) Missing/unknown293 (0.3)291 (0.3)2 (0.2)Alcohol intake<0.001 Never4,848 (4.2)4,821 (4.2)27 (3.4) Previous3,592 (3.1)3,541 (3.1)51 (6.4) Current106,425 (92.6)105,703 (92.6)722 (90.0) Missing/unknown116 (0.1)114 (0.1)2 (0.2)Physical activity<0.001 Low15,971 (13.9)15,845 (13.9)126 (15.7) Moderate40,176 (34.9)39,940 (35.0)236 (29.4) High35,223 (30.6)35,005 (30.7)218 (27.2) Missing/unknown23,611 (20.5)23,389 (20.5)222 (27.7)Diabetes0.001 No111,421 (96.9)110,660 (96.9)761 (94.9) Yes3,358 (2.9)3,321 (2.9)37 (4.6) Missing/unknown202 (0.2)198 (0.2)4 (0.5)Cardiovascular disease<0.001 No89,572 (77.9)89,094 (78.0)478 (59.6) Yes25,409 (22.1)25,085 (22.0)324 (40.4)Hypertension<0.001 No90,887 (79.0)90,315 (79.1)572 (71.3) Yes24,094 (21.0)23,864 (20.9)230 (28.7)Blood-pressure-lowering medication<0.001 No98,884 (86.0)98,259 (86.1)625 (77.9) Yes16,097 (14.0)15,920 (13.9)177 (22.1)Lipid-lowering medication<0.001 No101,758 (88.5)101,119 (88.6)639 (79.7) Yes13,223 (11.5)13,060 (11.4)163 (20.3)Insulin use0.189 No114,255 (99.4)113,461 (99.4)794 (99.0) Yes726 (0.6)718 (0.6)8 (1.0) Total energy intake, kJ/day8,022.71 ± 2,209.318,022.68 ± 2,207.948,028.12 ± 2,398.430.949 Sulfur microbial diet z-score0.00 ± 1.00−0.00 ± 1.000.04 ± 1.060.273Sulfur microbial diet quartile0.275 Q128,810 (25.1)28,613 (25.1)197 (24.6) Q228,705 (25.0)28,520 (25.0)185 (23.1) Q328,740 (25.0)28,543 (25.0)197 (24.6) Q428,726 (25.0)28,503 (25.0)223 (27.8)Notes: Bold P values indicate statistically significant differences (P < 0.05) between participants with incident lung cancer and participants without lung cancer. Baseline Characteristics According to Incident Lung Cancer Status Notes: Bold P values indicate statistically significant differences (P < 0.05) between participants with incident lung cancer and participants without lung cancer. Sulfur Microbial Diet Score and Lung Cancer Risk During follow-up, 802 incident lung cancer cases were documented. In the fully adjusted model (Model 3), women in the highest quartile of the SMD score had a significantly higher risk of lung cancer than those in the lowest quartile (HR 1.25, 95% CI 1.02–1.54), with a significant positive trend across quartiles (P for trend = 0.036). Each 1-SD increase in the SMD score was associated with an 8% higher risk of lung cancer (HR 1.08, 95% CI 1.00–1.16), an estimate of borderline statistical significance (P = 0.054). The corresponding results for Models 1 to 3 are shown in Table 3.Table 3Association Between the Sulfur Microbial Diet Score and the Risk of Lung CancerExposureModel 1HR (95% CI)PModel 2HR (95% CI)PModel 3HR (95% CI)PQ2 vs Q11.21 (0.98, 1.48)0.0751.18 (0.96, 1.45)0.1101.19 (0.97, 1.47)0.097Q3 vs Q11.28 (1.05, 1.57)0.0161.19 (0.97, 1.46)0.0951.20 (0.98, 1.48)0.079Q4 vs Q11.49 (1.22, 1.82)<0.0011.23 (1.01, 1.51)0.0441.25 (1.02, 1.54)0.034Per 1-SD increase————1.08 (1.00, 1.16)0.054P for trend (Model 3)0.036Notes: Model 1 was adjusted for age and ethnicity. Model 2 was further adjusted for body mass index, Townsend deprivation index, education, income, employment, smoking status, alcohol intake, physical activity, diabetes, cardiovascular disease and hypertension. Model 3 was further adjusted for blood-pressure-lowering medication, lipid-lowering medication, insulin use and total energy intake. P for trend was calculated by assigning the median value of each quartile and modelling it as a continuous variable. Q1–Q4, quartiles of the SMD score.Abbreviations: SMD, sulfur microbial diet; HR, hazard ratio; CI, confidence interval; SD, standard deviation. Table 3Association Between the Sulfur Microbial Diet Score and the Risk of Lung CancerExposureModel 1HR (95% CI)PModel 2HR (95% CI)PModel 3HR (95% CI)PQ2 vs Q11.21 (0.98, 1.48)0.0751.18 (0.96, 1.45)0.1101.19 (0.97, 1.47)0.097Q3 vs Q11.28 (1.05, 1.57)0.0161.19 (0.97, 1.46)0.0951.20 (0.98, 1.48)0.079Q4 vs Q11.49 (1.22, 1.82)<0.0011.23 (1.01, 1.51)0.0441.25 (1.02, 1.54)0.034Per 1-SD increase————1.08 (1.00, 1.16)0.054P for trend (Model 3)0.036Notes: Model 1 was adjusted for age and ethnicity. Model 2 was further adjusted for body mass index, Townsend deprivation index, education, income, employment, smoking status, alcohol intake, physical activity, diabetes, cardiovascular disease and hypertension. Model 3 was further adjusted for blood-pressure-lowering medication, lipid-lowering medication, insulin use and total energy intake. P for trend was calculated by assigning the median value of each quartile and modelling it as a continuous variable. Q1–Q4, quartiles of the SMD score.Abbreviations: SMD, sulfur microbial diet; HR, hazard ratio; CI, confidence interval; SD, standard deviation. Association Between the Sulfur Microbial Diet Score and the Risk of Lung Cancer Notes: Model 1 was adjusted for age and ethnicity. Model 2 was further adjusted for body mass index, Townsend deprivation index, education, income, employment, smoking status, alcohol intake, physical activity, diabetes, cardiovascular disease and hypertension. Model 3 was further adjusted for blood-pressure-lowering medication, lipid-lowering medication, insulin use and total energy intake. P for trend was calculated by assigning the median value of each quartile and modelling it as a continuous variable. Q1–Q4, quartiles of the SMD score. Abbreviations: SMD, sulfur microbial diet; HR, hazard ratio; CI, confidence interval; SD, standard deviation. Dose–Response Relationship Restricted cubic spline analysis indicated a positive and approximately linear dose–response association between the SMD score and lung cancer risk, with no evidence of departure from linearity (P for non-linearity = 0.396); the risk of lung cancer increased gradually across the range of higher SMD scores (Figure 2). Figure 2Restricted cubic spline analysis of the dose–response association between the sulfur microbial diet score and the risk of lung cancer in women. The solid line denotes the hazard ratio and the shaded area the 95% confidence interval.A line graph of hazard ratio versus sulfur microbial diet score, rising slightly at higher scores. Restricted cubic spline analysis of the dose–response association between the sulfur microbial diet score and the risk of lung cancer in women. The solid line denotes the hazard ratio and the shaded area the 95% confidence interval. Subgroup and Sensitivity Analyses The positive association between the SMD score and lung cancer risk was broadly consistent across subgroups defined by age, ethnicity, alcohol intake and smoking status, with no significant interaction in any subgroup (all P for interaction > 0.40); although the subgroup-specific estimates were nominally statistically significant among White women, current or previous smokers, and women consuming alcohol less than three times per week, these findings should be interpreted cautiously because none of the interaction tests was statistically significant (Supplementary Figure S1 and Supplementary Tables S2 and S3). In sensitivity analyses, the association was attenuated but directionally consistent after excluding lung cancer events within the first two years of follow-up (HR 1.22, 95% CI 0.99–1.51) and was strengthened after excluding women with diabetes or CVD at baseline (HR 1.44, 95% CI 1.12–1.84) (Supplementary Figure S2 and Supplementary Table S1). Substitution Analyses In substitution analyses, replacing processed meat with an equivalent amount of vegetables was associated with a lower risk of lung cancer, although the association did not reach statistical significance (HR 0.93, 95% CI 0.86–1.01; P = 0.076). Although the point estimate was consistent with a lower risk of lung cancer, the confidence interval included the null value. No appreciable associations were observed for the other modelled substitutions, including replacing sweets and desserts with vegetables or low-calorie drinks, and replacing low-calorie drinks with vegetables (Table 4).Table 4Substitution Analyses of Sulfur Microbial Diet Components and the Risk of Lung CancerSubstitutionHR (95% CI)P valueVegetables replacing processed meat0.93 (0.86, 1.01)0.076Vegetables replacing sweets and desserts1.00 (0.96, 1.05)0.892Low-calorie drinks replacing sweets and desserts0.99 (0.89, 1.11)0.907Vegetables replacing low-calorie drinks1.01 (0.90, 1.13)0.858Notes: Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional-hazards models adjusted as in Model 3 and pooled across multiply imputed datasets. Substitution models estimated the effect of replacing one dietary component with an equivalent amount of another while holding total intake constant. Table 4Substitution Analyses of Sulfur Microbial Diet Components and the Risk of Lung CancerSubstitutionHR (95% CI)P valueVegetables replacing processed meat0.93 (0.86, 1.01)0.076Vegetables replacing sweets and desserts1.00 (0.96, 1.05)0.892Low-calorie drinks replacing sweets and desserts0.99 (0.89, 1.11)0.907Vegetables replacing low-calorie drinks1.01 (0.90, 1.13)0.858Notes: Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional-hazards models adjusted as in Model 3 and pooled across multiply imputed datasets. Substitution models estimated the effect of replacing one dietary component with an equivalent amount of another while holding total intake constant. Substitution Analyses of Sulfur Microbial Diet Components and the Risk of Lung Cancer Notes: Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional-hazards models adjusted as in Model 3 and pooled across multiply imputed datasets. Substitution models estimated the effect of replacing one dietary component with an equivalent amount of another while holding total intake constant. Mediation Analyses Approximate mediation analyses indicated that circulating metabolic and inflammatory biomarkers explained little of the association between the SMD score and lung cancer risk. The estimated proportions mediated were small for triglycerides (1.71%), C-reactive protein (1.60%), HDL cholesterol (0.71%), diastolic blood pressure (0.10%) and total cholesterol (0.08%). Negative estimates were observed for BMI (−7.49%), systolic blood pressure (−0.60%) and glucose (−0.04%), indicating inconsistent or suppression-type mediation rather than a substantive mediating pathway. Overall, none of the evaluated biomarkers appeared to explain a meaningful proportion of the association (Table 5 and Figure 3).Table 5Approximate Mediation Analyses of the Association Between the Sulfur Microbial Diet Score and the Risk of Lung CancerMediatorTotal EffectHR (95% CI)PDirect EffectHR (95% CI)PMediatorEffect (95% CI)PApprox. %MediatedBMI1.04 (1.00, 1.08)0.0711.04 (1.00, 1.09)0.0540.99 (0.97, 1.00)0.132−7.49Glucose1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0541.02 (0.94, 1.10)0.646−0.04HDL cholesterol1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0550.75 (0.59, 0.96)0.0250.71Total cholesterol1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0540.98 (0.91, 1.04)0.4790.08Triglycerides1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0581.01 (1.00, 1.03)0.0311.71Systolic blood pressure1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0521.00 (1.00, 1.01)0.556−0.60Diastolic blood pressure1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0541.00 (0.99, 1.01)0.8950.10C-reactive protein1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0581.02 (1.00, 1.03)0.0181.60Notes: The total effect denotes the overall association between the sulfur microbial diet score and lung cancer risk; the mediator effect and direct effect decompose this association for each candidate mediator. A negative approximate percentage mediated indicates inconsistent (suppression-type) mediation. Each row corresponds to a separate single-mediator Cox regression model.Abbreviations: BMI, body mass index; HDL, high-density lipoprotein; HR, hazard ratio; CI, confidence interval. Figure 3Exploratory mediation analysis of the association between the sulfur microbial diet score and the risk of lung cancer. (A) shows the mediation framework. (B) presents the approximate proportion mediated by each candidate metabolic or inflammatory biomarker. Negative values indicate inconsistent (suppression-type) mediation.A diagram labeled ′A Mediation framework′ shows the relationship between SMD score, candidate biomarker and incident lung cancer. It includes paths labeled ′a path′ and ′b path′, with a direct effect labeled ′c prime′. The total effect is described as the sum of direct and indirect effects. Below, a bar chart labeled ′B Estimated mediation proportions′ displays estimated proportions mediated by various biomarkers. Negative values indicate suppression-type mediation. The chart lists BMI at negative 7.49 percent, glucose at negative 0.04 percent, HDL cholesterol at 0.71 percent, total cholesterol at 0.08 percent, triglycerides at 1.71 percent, systolic blood pressure at negative 0.60 percent, diastolic blood pressure at 0.10 percent and C-reactive protein at 1.60 percent. Positive mediation estimates are marked with blue circles and suppression-type estimates with orange circles.A diagram and bar chart of mediation analysis for SMD score and lung cancer risk.Abbreviations: BMI, body mass index; HDL, high-density lipoprotein; SMD, sulfur microbial diet. Table 5Approximate Mediation Analyses of the Association Between the Sulfur Microbial Diet Score and the Risk of Lung CancerMediatorTotal EffectHR (95% CI)PDirect EffectHR (95% CI)PMediatorEffect (95% CI)PApprox. %MediatedBMI1.04 (1.00, 1.08)0.0711.04 (1.00, 1.09)0.0540.99 (0.97, 1.00)0.132−7.49Glucose1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0541.02 (0.94, 1.10)0.646−0.04HDL cholesterol1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0550.75 (0.59, 0.96)0.0250.71Total cholesterol1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0540.98 (0.91, 1.04)0.4790.08Triglycerides1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0581.01 (1.00, 1.03)0.0311.71Systolic blood pressure1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0521.00 (1.00, 1.01)0.556−0.60Diastolic blood pressure1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0541.00 (0.99, 1.01)0.8950.10C-reactive protein1.04 (1.00, 1.09)0.0541.04 (1.00, 1.09)0.0581.02 (1.00, 1.03)0.0181.60Notes: The total effect denotes the overall association between the sulfur microbial diet score and lung cancer risk; the mediator effect and direct effect decompose this association for each candidate mediator. A negative approximate percentage mediated indicates inconsistent (suppression-type) mediation. Each row corresponds to a separate single-mediator Cox regression model.Abbreviations: BMI, body mass index; HDL, high-density lipoprotein; HR, hazard ratio; CI, confidence interval. Approximate Mediation Analyses of the Association Between the Sulfur Microbial Diet Score and the Risk of Lung Cancer Notes: The total effect denotes the overall association between the sulfur microbial diet score and lung cancer risk; the mediator effect and direct effect decompose this association for each candidate mediator. A negative approximate percentage mediated indicates inconsistent (suppression-type) mediation. Each row corresponds to a separate single-mediator Cox regression model. Abbreviations: BMI, body mass index; HDL, high-density lipoprotein; HR, hazard ratio; CI, confidence interval. Exploratory mediation analysis of the association between the sulfur microbial diet score and the risk of lung cancer. (A) shows the mediation framework. (B) presents the approximate proportion mediated by each candidate metabolic or inflammatory biomarker. Negative values indicate inconsistent (suppression-type) mediation. Discussion In this prospective analysis of 114,981 women, greater adherence to the sulfur microbial diet was associated with a modestly higher risk of incident lung cancer. Women in the highest quartile of the SMD score had a 25% higher risk of lung cancer than those in the lowest quartile (HR 1.25, 95% CI 1.02–1.54), with a significant positive trend across quartiles, and the dose–response relationship was positive and approximately linear. The association was consistent in direction across subgroups defined by age, ethnicity, alcohol intake and smoking status, and it persisted in sensitivity analyses that addressed reverse causation and pre-existing cardiometabolic disease. Substitution analyses suggested that replacing processed meat with vegetables may be associated with lower risk, whereas circulating metabolic and inflammatory biomarkers explained little of the observed association. These findings are consistent with, and extend, a growing literature linking the SMD to adverse health outcomes. Since the dietary pattern was first derived as a correlate of faecal sulfur-metabolizing bacteria,21 greater adherence to the SMD has been associated with an increased risk of distal colorectal cancer in men,21 colorectal cancer overall and in both sexes,22 early-onset colorectal cancer precursors in women,23 colorectal cancer of specific molecular subtypes24 and colorectal adenoma among older adults.25 More recently, a population-based cohort reported that higher SMD adherence was associated with greater all-cause and cancer mortality.26 The magnitude of the association we observed for lung cancer (highest versus lowest quartile HR 1.25) is broadly comparable to that reported for colorectal cancer (highest versus lowest category HR approximately 1.27).22 To our knowledge, however, the present study is the first to examine the SMD in relation to an extra-intestinal solid tumour and the first to do so in a cohort restricted to women, suggesting that the health relevance of a sulfur-microbial-promoting diet may extend beyond the gastrointestinal tract. Our results are also coherent with the broader literature on diet and lung cancer. The SMD is characterised by higher consumption of processed meat and lower consumption of vegetables and legumes, and each of these components has been independently associated with lung cancer risk. In the same cohort, higher intakes of red and processed meat were associated with greater lung cancer risk, whereas fruit, vegetable and fibre intake and adherence to a “prudent” dietary pattern were associated with lower risk;9 meta-analyses of observational studies and pooled cohort analyses have reached similar conclusions.8,10 By integrating several of these foods into a single, microbially anchored index, the SMD may capture a combination of unfavourable dietary exposures more completely than any single food group. The substitution analysis reinforces this interpretation: replacing processed meat—an IARC-classified carcinogen11 whose adverse effects have been linked to N-nitroso compounds and other contaminants12—with vegetables was associated with a lower, albeit not statistically significant, risk of lung cancer, a direction of effect concordant with dietary recommendations that favour plant-based foods over processed meat. In substitution analysis, replacing processed meat with vegetables yielded a borderline non-significant inverse association with lung cancer risk (HR 0.93, P=0.076). Several factors may jointly account for the absence of statistical significance. First, the sample size of lung cancer cases may not provide adequate statistical power to detect a modest protective effect of vegetable substitution. Second, residual confounding from unmeasured environmental and occupational exposures cannot be fully eliminated, which may dilute the true protective signal of plant-based foods. Third, this is an isocaloric theoretical substitution model rather than an actual dietary intervention; real-world simultaneous changes in multiple food groups and long-term cumulative dietary intake cannot be fully captured by this single-item substitution approach. Biologically, processed meat contains multiple carcinogenic contaminants, while vegetables supply anti-inflammatory and antioxidant phytochemicals that suppress gut-derived toxic metabolites such as hydrogen sulfide. The consistent directional trend aligns with established lung cancer dietary guidelines favouring vegetable intake over processed meat, even though the association did not cross the conventional P<0.05 threshold in the present analysis. Future large-scale interventional or longer-term cohort studies are needed to further verify this protective substitution effect. Several mechanisms could plausibly underlie the observed association. The SMD was constructed to track the abundance of sulfur-metabolizing bacteria, which reduce dietary sulfur to hydrogen sulfide; at the concentrations found in the gut, hydrogen sulfide is genotoxic to epithelial cells, impairs DNA repair and promotes mucosal inflammation.18–20 Although these effects have been characterised principally in the colon, the gut microbiome exerts systemic effects: through the gut–lung axis, microbial dysbiosis and the metabolites and immune mediators it generates can modulate pulmonary immune surveillance and inflammation, processes increasingly implicated in lung carcinogenesis.15,17 Healthy dietary patterns, by contrast, favourably modulate gut microbial composition and lower systemic inflammation,35 and hydrogen-sulfide-related signalling has been shown to promote tumour proliferation, angiogenesis and metastasis in experimental models.28 In our data, however, conventional circulating metabolic and inflammatory biomarkers—including C-reactive protein—explained only a negligible fraction of the association, with the only statistically significant mediator being HDL cholesterol, which mediated merely 0.71% of the overall association and thus bears little biological relevance. We also observed mild suppression effects for several cardiometabolic biomarkers, meaning these factors slightly mask the crude association between SMD and lung cancer. This pattern suggests that, if the association is causal, it is unlikely to operate predominantly through the metabolic and inflammatory markers measured here, and may instead involve microbial, immunological or local pulmonary pathways that were not captured. We also observed mild suppression effects for several cardiometabolic biomarkers, meaning these factors slightly mask the crude association between SMD and lung cancer. The persistence of a positive association among never-smokers in subgroup analysis is consistent with a tobacco-independent contribution, although the relatively small number of events in that stratum warrants caution. From a public-health perspective, the SMD is composed entirely of modifiable dietary components, and the direction of our findings aligns with established dietary guidance favouring vegetables, legumes and whole foods over processed meat and sugar-sweetened or low-calorie beverages. Although the relative effect size was modest, dietary patterns are highly prevalent exposures, and even small relative risks can translate into a meaningful number of cases at the population level, particularly as the burden of lung cancer in women continues to rise.1,3 Our decision to study women specifically is supported by the distinct epidemiology of lung cancer in women, including the high proportion of cases among never-smokers and the potential influence of the hormonal environment on lung cancer development.4,7,29 These results identify dietary pattern as a candidate modifiable factor that, if confirmed, could complement tobacco control in lung cancer prevention among women. This study has several strengths. It was based on a large, well-characterised prospective cohort with linkage to national cancer registries, which minimised outcome misclassification and loss to follow-up. Dietary intake was assessed using repeated 24-hour dietary recalls and averaged across assessments, reducing measurement error relative to a single recall. We applied multiple imputation to limit bias from missing covariate data, adjusted for a broad range of sociodemographic, lifestyle and clinical confounders, and examined the robustness of the findings through pre-specified subgroup, sensitivity, substitution and mediation analyses. Finally, by focusing exclusively on women, the study addresses a population in which the epidemiology of lung cancer is distinct and for which evidence on modifiable dietary risk factors remains limited. Several limitations should also be acknowledged. First, the observational design precludes causal inference, and despite extensive adjustment, residual or unmeasured confounding—particularly by smoking, the dominant cause of lung cancer—cannot be excluded; reassuringly, the association was directionally consistent across smoking strata and persisted after excluding early events. Second, the SMD score was derived from dietary patterns identified in North American health-professional cohorts and was not validated against direct measurement of the gut microbiome; the score therefore represents a dietary proxy rather than a measured microbial phenotype. Third, diet was assessed only at baseline and may not reflect long-term intake or changes over follow-up. Fourth, the number of incident lung cancer cases, although adequate for the primary analysis, limited statistical power in some subgroups, and the per-1-SD association was of borderline significance. Fifth, neither the gut microbiome nor hydrogen sulfide was measured directly, so the proposed biological pathway remains hypothetical. Sixth, participants are predominantly of White ethnicity and are healthier and less socioeconomically deprived than the general population, which may limit the generalisability of the absolute estimates, although internal exposure–outcome associations are likely to remain broadly valid.36 Finally, baseline data lacked standardized indicators of passive smoking, long-term air pollutant exposure, occupational carcinogen exposure, menopausal status, family history of lung cancer, asthma, chronic bronchitis, emphysema and COPD; these critical lung cancer-related factors could not be incorporated as covariates in regression models. Unmeasured confounding stemming from these uncollected variables may slightly distort the association between SMD and lung cancer risk. However, the positive association between SMD and lung cancer remained consistent across stratified analyses by smoking status (never, former, current smokers), which partially alleviates concerns over residual confounding related to tobacco exposure. Future cohort studies equipped with comprehensive environmental, occupational and female-specific health records are required to replicate and validate our findings. Conclusion In this large prospective cohort of women, greater adherence to the sulfur microbial diet—a dietary pattern linked to the abundance of sulfur-metabolizing gut bacteria—was associated with a modestly higher risk of incident lung cancer, with a positive and approximately linear dose–response relationship. The association was robust across subgroup and sensitivity analyses and was not appreciably explained by conventional metabolic or inflammatory biomarkers, while substitution analyses suggested that replacing processed meat with vegetables may be beneficial. These findings extend the relevance of the sulfur microbial diet beyond colorectal neoplasia to lung cancer and identify diet as a candidate modifiable factor in lung cancer prevention among women. Further studies that incorporate direct measurement of the gut microbiome and that replicate these findings in diverse populations are warranted to clarify the underlying mechanisms and to establish causality. For future research directions, several promising areas deserve further exploration. First, cohort studies with direct gut microbiome and hydrogen sulfide metabolite detection are required to validate the gut–lung mechanistic pathway linking SMD to lung cancer. Second, replication analyses in multi-ethnic populations and cohorts with complete records of passive smoking, air pollution and occupational toxicants are needed to reduce residual confounding and improve external validity. Third, dietary intervention trials are necessary to verify whether replacing processed meat with vegetables can effectively lower lung cancer risk. Additionally, sex-specific mechanistic research is warranted to reveal hormonal interactions between SMD and lung tumour progression in female populations. Collectively, the present work lays an epidemiological foundation for subsequent basic, translational and population-based research on diet-gut microbiota and female lung cancer.