This study utilized data from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. The PLCO trial is a large randomized controlled trial that was designed to evaluate the effectiveness of cancer screening tests for reducing cancer mortality rates. This trial was conducted between 1993 and 2001 at 10 clinical centers in the United States and enrolled 154,887 participants aged 55–74. All participants were randomly assigned to either a control group or an intervention group involving screening tests. The follow-up period extended until 2009 for the incidence of over 20 types of cancer, including HNC, and until 2018 for cancer-related mortality. The PLCO trial extensively collected data on the demographic characteristics, health history, lifestyle factors, and diet information of the participants through self-reported questionnaires. In this trial, participants were asked to complete two questionnaires: the baseline questionnaire (BQ) and the diet history questionnaire (DHQ) at the beginning of the trial. The DHQ relied on a 137-item food frequency questionnaire (FFQ) to gather data on dietary information over the past year, and the DRRD dietary pattern can be well established using the dietary data collected through DHQ. Detailed information on the PLCO trial has been reported in related literature (13).

The objective of our current study was to investigate whether adherence to the DRRD is related to the risk of HNC. The primary endpoint was defined as the diagnosis of HNC among participants, and the follow-up time was determined as the period from DHQ completion to the occurrence of HNC, death, loss during follow-up, or the end of the follow-up period (i.e., December 31, 2009), whichever occurred first (Figure 1). To achieve the study objective, a set of exclusion criteria were employed to establish an appropriate study cohort from an initial pool of 154,887 participants. Firstly, 4,918 participants who did not return the BQ were excluded. Secondly, 38,463 participants who either did not return the DHQ or returned an incomplete DHQ that having at least 8 missing frequency responses of dietary items, a missing completion date, completion date after death, or extreme energy consumption (top 1% and bottom 1%) were excluded. Thirdly, 9,683 participants with a history of any cancer prior to DHQ entry were excluded. Fourthly, 68 participants who experienced an outcome event between DHQ entry and DHQ completion were excluded. Ultimately, the remaining cohort comprised 101,755 individuals in our study, as illustrated in Figure 2.

To evaluate the adherence of each participant to DRRD, a DRRD score was calculated based on the methodology described in previous studies (9). Briefly, the intakes of the nine DRRD components were obtained from the DHQ, and then each component was categorized into five groups based on its quintile values of dietary intake, and assigned scores ranging from 1 to 5. For cereal fiber, coffee, nuts, whole fruits, and the ratio of polyunsaturated to saturated fat, a higher quintile value indicated a higher score. Conversely, for trans-fat, GI, SSBs, and red and processed meats, a lower quintile value indicated a higher score. The DRRD score was then calculated by summing the scores of the nine components, resulting in a range of 9 to 45. An increased DRRD score indicates greater adherence to the DRRD dietary pattern. Detailed data for determining DRRD score was shown in Supplementary Table S1.

The ascertainment of HNC cases primarily relied on the administration of an annual study update form, which was disseminated by each screening center to participants. This form was designed to elicit information on whether individuals had received a diagnosis of HNC, along with the date and location of the diagnosis, and the contact details of their healthcare providers. HNC cases were defined based on the following ICD-O-2 codes for malignant tumors: (1) oral cavity: C00.3–C00.9, C02.0–C02.3, C03.0, C03.1, C03.9, C04.0, C04.1, C04.8, C04.9, C05.0, C06.0–C06.2, C06.8 and C06.9; (2) oropharynx: C01.9, C02.4, C05.1, C05.2, C09.0, C09.1, C09.8, C09.9, C10.0, C10.2–C10.4, C10.8 and C10.9; (3) hypopharynx: C12.9, C13.0–C13.2, C13.8 and C13.9; (4) oral cavity or pharynx NOS: C02.8, C02.9, C05.8, C05.9, C14.0, C14.2 and C14.8; and (5) larynx: C10.1, C32.0–C32.3 and C32.8–C32.9. Cases of NHC reported through this form were subjected to further verification by scrutinizing any available medical records. In addition, supplementary sources such as death certificates and family reports were utilized to augment the ascertainment process. To ensure consistency in case selection, only participants who had received a diagnosis of HNC were included in the study.

The study gathered information on age at DHQ completion, drinking status and alcohol consumption, energy intake, food and nutrient consumption via the DHQ. Daily food intake was determined by multiplying food frequency by portion size, while daily nutrient intake was estimated using the USDA’s 1994–96 Continuing Survey of Food Intakes by Individuals and the Nutrition Data Systems for Research (14). The detailed calculation methods of dietary fiber, GI, and trans-fat are available in previously published literatures (15–17). Additional data or covariates, such as gender, race, body mass index (BMI), educational level, marital status, smoking status, pack-years of smoking, history of diabetes, and family history of HNC, were obtained using a self-administered baseline questionnaire.

To reduce potential biases and enhance the statistical power of our study, imputation was performed using modal values for categorical variables and median values for continuous variables. Supplementary Table S2 displays the distribution of variables with missing values before and after imputation. To examine the potential impact of data imputation on our results, we also repeated the primary statistical analyses in the population with complete covariate data in the subsequent sensitivity analysis.

To evaluate the association between DRRD and HNC risk, the study employed Cox proportional hazards regression as the primary analysis model, with follow-up period as the time metric. The DRRD score was categorized into quartiles, with the lower quartile serving as the reference group. Person-years of each quartile were estimated based on the duration of follow-up. To assess whether a linear trend could be observed across quartiles of DRRD scores for estimating HNC risk, median values of each quartile were assigned to individuals within the corresponding quartile and treated as a continuous variable in regression models. After examining the Schoenfeld residuals, we found that the proportional hazards assumption of the Cox regression model was satisfied (P for global test > 0.05). Multivariable regression models were utilized to further adjust potential covariates. Specifically, model 1 was adjusted for age, sex, and race. Model 2 included additional adjustments for marital status, educational level, BMI, family history of HNC, smoking status, pack-years of smoking, drinking status, alcohol consumption, history of diabetes, and energy from diet. A restricted cubic spline model with three knots (i.e., 10th, 50th, and 90th percentiles of DRRD score) was employed to analyze HNC risk across the entire range of the DRRD score. Additionally, we conducted further analyses to investigate the association between the nine dietary components of the DRRD and HNC risk using similar Cox regression model as described above. Specifically, we obtained the intake of each dietary component of the DRRD from the DHQ and divided them into quartiles, with the lowest quartile serving as the reference group.

To investigate whether the association between DRRD score and HNC risk was modified by various factors, subgroup analyses were conducted. Participants were divided into categories based on age (>65 vs. ≤65 years), sex (male vs. female), BMI (≤25 vs. >25 kg/m2), smoking status (never vs. current or former), pack-years of smoking (≤medium vs. >medium), drinking status (no vs. yes), alcohol consumption (≤medium vs. >medium), history of diabetes (no vs. yes), Family history of HNC (no vs. yes), and energy from diet (≤medium vs. >medium). Interaction p values were computed by comparing models with and without multiplicative interaction terms before subgroup analyses to avoid spurious subgroup effects. Additionally, sensitivity analyses were conducted to confirm the robustness of the primary results. These included repeating the primary analysis in participants with complete data, excluding participants with diabetes, excluding participants with follow-up less than 2 years, excluding participants with extreme energy intake (>4,000 kcal/day or <500 kcal/day), and excluding participants with extreme BMI (top 1% and bottom 1%).

The statistical significance level was set at a p value of < 0.05. R 4.2.1 software was utilized for all statistical analyses.

In this study, a total of 101,755 individuals were included and categorized into quartiles based on their DRRD scores: quartile 1 (n = 27,890) with scores between 9 and 23, quartile 2 (n = 28,970) with scores between 24 and 27, quartile 3 (n = 19,784) with scores between 28 and 30, and quartile 4 (n = 25,111) with scores between 31 and 45. The mean (standard deviation) DRRD score for all participants was 26.84 (5.31), and their baseline characteristics were presented in Table 1. Compared to the lowest quartile group, individuals in the highest quartile group tended to be female, older, have a lower BMI, non-smoker or have fewer pack-years of smoking, a drinker or have high alcohol consumption, and have no history of diabetes. Moreover, those in the highest quartile of DRRD scores had a lower intake of energy compared to those in the lowest quartile.

This study followed up with a total of 900001.9 person-years and recorded 279 cases of malignant primary HNC. The overall incidence was 3.1 cases/10,000 person-years with a mean (standard deviation) follow-up duration of 8.84 (1.92) years. The results of Cox regression analysis of the entire study population are presented in Table 2. The unadjusted model analysis showed that individuals in the highest quartile group had a lower risk of HNC compared to those in the lowest quartile group (HR_{Q4 vs. Q1}: 0.392; 95% CI: 0.271, 0.568; p < 0.001 for trend). After adjusting for all potential confounding factors, the inverse association between DRRD score and HNC risk remained significant (HR_{Q4 vs. Q1}: 0.582; 95% CI: 0.396, 0.856; p = 0.005 for trend). In the restricted cubic spline regression model, DRRD score was found to have an inverse association with the risk of HNC in a linear dose-response manner (p = 0.211 for non-linearity), as shown in Figure 3.

The subgroup analyses results are presented in Table 3, indicating that the inverse correlation between DRRD score and HNC risk was not modified by various factors such as age, sex, BMI, pack-years of smoking, drinking status, alcohol consumption, history of diabetes and energy from diet (all P for interaction > 0.05). However, the inverse correlation was more pronounced among participants who had never smoked (HR_{Q4 vs. Q1}: 0.193; 95% CI: 0.073, 0.511; p < 0.001 for trend) compared to current or former smokers (p = 0.044 for interaction).

The results of several sensitivity analyses are presented in Table 4, and indicate that the inverse association between DRRD score and HNC risk did not change significantly, thereby further confirming the robustness of our primary findings. Specifically, when Cox regression analyses were repeated in participants with complete covariate data, we obtained similar results (HR_{Q4 vs. Q1}: 0.542; 95% CI: 0.360, 0.816; p = 0.003 for trend).

We further investigated the association between each dietary components of DRRD and the risk of HNC. Our Results indicated that individuals in the highest quartile of cereal fiber and whole fruit consumption, considered as “favorable” DRRD components, had a lower risk of HNC compared to those in the lowest quartile [(cereal fiber: HR_{Q4 vs. Q1}: 0.471; 95% CI: 0.301, 0.739; p = 0.002 for trend) (Supplementary Table S3) and (whole fruit: HR _{Q4 vs. Q1}: 0.555; 95% CI: 0.372, 0.829; p = 0.002 for trend) (Supplementary Table S4)]. However, there was no significant association between the risk of HNC and other DRRD components, such as nuts, coffee, polyunsaturated/saturated fatty acids, trans-fat, GI, SSBs, and red meat and processed meat (Supplementary Tables S5–S11).