This study prospectively examines participants from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a large-scale randomized clinical trial sponsored by the National Cancer Institute (NCI) from 1993 to 2001 (18). Approximately 150,000 men and women, aged 55–74 years, were recruited from ten screening centers nationwide. Participants were randomly allocated to either routine medical care (control group) or additional cancer screenings (intervention group) (19). The trial’s protocol received approval from institutional review boards at both the NCI and participating centers, with all participants providing informed, written consent. Detailed aspects of the PLCO trial design, including power calculations and recruitment methodologies, have been thoroughly documented in previous publications (20, 21). The PLCO trial protocol was approved by the NCI’s and participating centers’ institutional review boards, and all participants gave informed written consent.

Within the PLCO trial, baseline demographic and lifestyle data were gathered from participants using self-administered questionnaires (Baseline Questionnaire, BQ). The key variables analyzed in our study encompassed age, sex, race, marriage, education level, smoking status, daily cigarette consumption, body mass index (BMI) at baseline, history of aspirin use, diabetes, hypertension, colorectal diverticulitis/diverticulosis, colorectal polyp and colorectal comorbidities (specifically Gardner’s syndrome, ulcerative colitis, Crohn’s disease, or familial polyposis), and family history of CRC. BMI was computed as weight (kg) divided by height squared (m²). Dietary intake information was collected using a validated 137-item Food Frequency Questionnaire (FFQ), termed the Dietary History Questionnaire (DHQ), which was administered 3 years post-enrollment in the PLCO trial. The DHQ evaluated portion sizes, frequencies, and types of foods and supplements consumed by participants over the preceding year. The validity of the DHQ was established through comparison with a 24-h dietary recall study (i.e., the Eating at America’s Table Study) (22). In this study, the DHQ demonstrated superior performance in assessing absolute nutrient intake compared to other commonly utilized FFQs, such as the Block and Willett questionnaires (22).

To ensure robust and unbiased estimation of protein quality-CRC risk associations, participants were further excluded based on the following criteria:

Ultimately, as illustrated in Figure 1, a total of 101,709 participants met the inclusion criteria, comprising 52,250 females and 49,459 males.

Healthy plate protein quality index is established based on the nutritional protein quality recommendations derived from the latest international dietary guidelines (23, 24). To reflect differential protein bioavailability and disease associations, the numerator includes high-quality plant and animal protein sources (seafood, poultry, pulses, nuts) with favorable metabolic profiles, while the denominator comprises protein sources (red/processed meats, cheese) linked to increased CRC risk through mechanisms such as heme iron, saturated fats, and advanced glycation end products. The quality of dietary protein was determined by computing HPPQI using the following ratios (25):

In the PLCO trial, CRC case identification primarily relied on annual study update questionnaires mailed to surviving participants, which solicited information regarding any new cancer diagnoses. Self-reported CRC cases were subsequently validated through standardized medical record review, with study physicians adjudicating diagnoses in a blinded manner. Participant vital status was tracked using these annual questionnaires, with multiple follow-up attempts made to non-responders. Additional mortality surveillance involved routine linkages with the National Death Index and examination of death certificates, utilizing ICD-9 coding to ascertain causes of death.

Colorectal cancer cases in this study were classified by anatomic subsite according to International Classification of Diseases for Oncology (ICD-O2) codes, distinguishing between proximal colon (C180-C185), distal colon (C186-C187), and rectal (C199-C209) cancers. For subsite-specific analyses of colorectal cancers, cases coded as C188, C189, C212, and C218 were excluded from consideration.

In the present analysis, covariate data exhibited varying degrees of missingness. For categorical variables with <5% missingness, such as marital status, race, education, smoking status, daily cigarette consumption, aspirin use history, history of hypertension, diabetes, colorectal diverticulitis/diverticulosis, colorectal polyps and colorectal comorbidities, and family history of cancer/CRC, missing values were imputed using the mode. Continuous variables with <5% missing data, including BMI (both baseline and change) and pack-years of cigarette smoking, were imputed using the median (26). The detailed imputation information for each missing data item and its corresponding proportion is presented in Supplementary Table 1.

This study defined time-to-CRC-event as the number of days from DHQ completion to CRC diagnosis or CRC-related death. Follow-up for primary outcomes extended from DHQ completion until CRC diagnosis, death, loss to follow-up, or December 31, 2009 (the end of cancer incidence follow-up), whichever occurred first. Mortality follow-up for secondary outcomes continued until 2018, as detailed in Figure 2 on the PLCO website.¹ Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for associations between the HPPQI and outcomes, with follow-up duration serving as the time metric. HPPQI was categorized into quartiles, with a higher HPPQI category indicating better dietary protein quality. The first quartile of HPPQI was used as the referent. Additionally, continuous variables were created using quartile medians to assess linear trends, and significance was reported as p-values. Potential confounders included established CRC risk factors and clinically relevant variables (27). Cox models incorporated two adjustment sets: Model 1 adjusted for demographic factors (sex, age, race, education and marital status), and Model 2 further adjusted for lifestyle/clinical factors (BMI, smoking status, daily cigarette use, hypertension, diabetes, history of colorectal comorbidities, polyps and diverticulitis/diverticulosis, aspirin use and family history of CRC) and trial group. Restricted cubic spline (RCS) models were used to characterize non-linear associations between HPPQI and CRC incidence/mortality, using the median HPPQI as the reference, with non-linearity assessed by testing the regression coefficient of the second spline term against zero (28, 29). Identical analyses were conducted for CRC anatomical subsites. In addition, we employed Kaplan - Meier survival curves to describe the association between the HPPQI and the incidence/mortality rates of CRC. Moreover, we conducted a log - rank test to compare the differences in survival curves among different HPPQI level groups.

To assess potential effect modification of the relationship between HPPQI and CRC incidence and mortality across key factors, we conducted prespecified subgroup analyses. Subgroups were categorized into several categories: demographic characteristics (age > 65 vs. ≤65 years, sex male vs. female, race White vs. non-White, marital status married vs. unmarried), health conditions (diabetes status yes vs. no, hypertension status yes vs. no, baseline BMI ≤ 30 vs. >30 kg/m²), family and medical history (family history of CRC absent vs. present, history of colorectal diverticulitis/diverticulosis yes vs. no, history of colorectal comorbidities yes vs. no, history colorectal polyps yes vs. no), and lifestyle factors (smoking status never vs. current/former, aspirin use no vs. yes, daily cigarette consumption 0 vs. 1–20 vs. >20 cigarettes). To identify any potential spurious subgroup effects, interaction P-values were evaluated by comparing models with and without interaction terms.

To enhance the robustness of the findings, we conducted some sensitivity analyses (27, 30):

All statistical analyses were carried out using R software version 4.3.1, with two-tailed P < 0.05 as the level of statistical significance.

In this study, the median HPPQI for the participants was 0.91, with an interquartile range (IQR) of 1.13. Participants were categorized into four quartiles based on their HPPQI values: Quartile 1 (0–0.54), Quartile 2 (0.54–0.91), Quartile 3 (0.91–1.67), and Quartile 4 (>1.67). Table 1 shows that participants in Q4 (the highest quartile) were more likely to be female, non-white, and highly educated, but less likely to be married, users of aspirin, or be diagnosed with diabetes, hypertension, or polyps. Moreover, participants in Q4 are less likely to have a family history of cancer, have lower smoking intensity and have a lower and less variable BMI.

During the mean follow-up period of 8.82 years (896,103 person-years), a total number of 1100 CRC cases were reported, including 648 proximal colon cancer cases, 226 distal colon cancer cases and 204 rectal cancer cases, resulting in an overall incidence rate of approximately 12.28 cases per 10,000 person-years. As illustrated in Table 2, the Cox regression analysis revealed a significant negatively association between higher HPPQI values and CRC incidence after adjustment for potential confounding factors (HR Q4 vs. Q1: 0.77; 95% CI: 0.65, 0.93. p = 0.009 for trend). But in subsite analyses, HPPQI was not significantly associated with the incidence of proximal colon cancer, distal colon cancer or rectum cancer (all P > 0.05 for trend). The RCS model revealed a linear relationship between HPPQI and the risk of overall CRC incidence (Figure 3). The survival curve results indicate that there are significant differences in survival curves among different HPPQI level groups (P = 0.0051). As time progresses, the high - HPPQI group demonstrates a generally lower incidence of colorectal cancer overall (Supplementary Figure 1).

Stratified analyses assessed potential effect modification by participant characteristics (Supplementary Table 2), with the inverse association persisting across most subgroups defined by demographic characteristics (age, sex, race, marital status), health conditions (diabetes status, hypertension, baseline BMI), family and medical history (family history of CRC and history of colorectal comorbidities, polyps, and diverticulitis/diverticulosis), and lifestyle factors (aspirin use, smoking status, and daily cigarette consumption). However, none of the investigated factors emerged as a statistically significant moderator of the primary association based on tests for interaction. Additionally, the relationships between HPPQI and the incidence of CRC was robustly supported by sensitivity analyses (Supplementary Table 3). The survival curve results reveal significant differences in survival curves among different HPPQI level groups (P = 0.0018). As time progresses, the high - HPPQI group exhibits a generally lower mortality rate of colorectal cancer overall (Supplementary Figure 2).

During the mean follow-up period of 15.07 years (1,532,681 person-years), a total number of 314 deaths attributed to CRC were identified, including 184 proximal colon cancer cases, 71 distal colon cancer cases and 41 rectal cancer cases, resulting in an overall mortality rate of approximately 2.05 cases per 10,000 person-years. As illustrated in Table 3, the Cox regression analysis revealed a significant negatively association between higher HPPQI values and CRC mortality after adjustment for potential confounding factors (HR Q4 vs. Q1: 0.66; 95% CI: 0.47, 0.91. p = 0.024 for trend). But in subsite analyses, HPPQI was not significantly associated with the mortality of proximal colon cancer, distal colon cancer or rectum cancer (all P > 0.05 for trend). The RCS model revealed a non-linear relationship between HPPQI and the risk of overall CRC mortality (Figure 4).

Similar to the incidence-based findings, stratified analyses were conducted to evaluate the potential effect modification by participant characteristics (Supplementary Table 4). It was observed that the inverse association persisted across the majority of subgroups. Nevertheless, based on the tests for interaction, none of the investigated factors emerged as statistically significant moderators of the primary association. Sensitivity analyses demonstrated a significant robust negative association between HPPQI value and CRC mortality rates (Supplementary Table 5).