This study included patients from 15 institutions (academic cancer centers and community hospitals) participating in the Florida Pancreas Collaborative (FPC), a state-wide population-based prospective cohort study and biobank conducted between May 2018 and September 2021 (46). Participants were eligible if they: a) were 18 years old and above, b) self-reported as African American or Black people, NHW, or H/L [the populations having the greatest PDAC burden in Florida and nationally (20)], c) were willing to complete the study questionnaire and donate biospecimens at the time of standard-of-care (SOC) procedures, and d) had a strong suspicion of PDAC and were treatment-naïve. The diagnosis was confirmed by pathologic review of tissue obtained through routine diagnostic procedures. The study was approved by the Advarra Institutional Review Board, and all patients provided informed consent to participate.

As described in a whitepaper from the FPC (46), data collection occurred at baseline/enrollment and at 6- and 12-months post-baseline via health screens, questionnaires, and case report forms. The 3-page health screen comprises the abridged version of the Patient-Generated Subjective Global Assessment (aPG-SGA), a revised version of the Edmonton Symptom Assessment System (ESAS-r), and the Canadian Problem Checklist ( Supplementary Figures 1A-C ). The aPG-SGA (47) is a four-part questionnaire that gathers self-reported information as part of “boxes” pertaining to: 1) height, weight, and weight change over the past two weeks, 2) food intake over the past month, 3) presence of symptoms affecting food intake over the past two weeks, and 4) activities and function over the past month. ESAS-r assesses the prevalence and severity of symptoms (e.g., pain, tiredness/fatigue) on a scale of 0 (“no symptoms”) to 10 (“worst symptoms”) (48). The Canadian Problem Checklist (CPC) is based on supportive care needs of cancer patients and includes domains of emotional, informational, practical, spiritual, social, family, and physical concerns (49, 50). Information on cigarette smoking over a lifetime and the past 30 days was also ascertained in the health screen.

Participants also completed a self-administered web- or teleform-based baseline questionnaire and abbreviated versions at follow-up time-points. The baseline questionnaire gathered demographic, socioeconomic, epidemiologic, and clinical variables. We also assessed self-reported grip strength via questionnaire since using a dynamometer was cost-prohibitive for our multi-center study (51). Two questions were asked: “Have you found it challenging to grip objects with your entire hand (such as when giving a hand-shake or when holding on to a stairwell or carrying a handle of a bucket)?” and “Have you found it challenging to grip objects with your fingers (such as a piece of paper or a coin)?” Responses included not at all, a little, quite a bit, and very much. Validated questionnaires were used to assess mental health, lifetime stress, nutrition, and QoL (52–54). Performance status was assessed at the clinic visit by treating providers using ECOG guidelines (55). Data was obtained from the electronic medical record (EMR) on presenting symptoms, anthropometric measures, comorbidities (56), primary tumor location (head, body, tail, other), imaging studies, surgical and pathology details, chemotherapy regimen, and lab values. Treatment and follow-up data from the Florida Cancer Data System (57) was also obtained. Vital status was verified using the National Death Index, Social Security Death Index, obituaries, autopsy reports, and death certificates.

Most participants had routine blood analyses performed as part of SOC, and had data on pre-treatment laboratory values considered in the Vigano classification (15) (CRP, albumin (Alb), hemoglobin (HgB), and white blood cell (WBC) count). Additionally, based on data supporting serum cancer antigen 19–9, the Glasgow Prognostic Score (GPS) (objective scoring based on concentrations of CRP and albumin), neutrophil to lymphocyte ratio (NLR), platelet count, and bilirubin as having prognostic value for patients with PDAC and other cancer types (58–65), these markers were also considered using previously-defined cut-offs (66–70). Since CRP was not routinely ordered as part of SOC for all participants, serum levels of CRP were evaluated as part of a multiplexed array research panel ( Supplementary Methods ) with other SOC markers and novel biomarkers of interest that are not a focus of Vigano (15) using archived serum from a subset of cases. Since biomarker concentrations may be influenced by antibiotics, steroids, or chemotherapy (27, 67, 71), neo- or adjuvant therapy and medication details were considered.

Participants were classified into four CC stages (NCa, PCa, Ca, and RCa) at baseline using criteria in Table 1 : abnormal biochemistry (A), decreased food intake (B), minimal weight loss (C), significant weight loss (D), and decreased activities and functioning (E-F). Due to small numbers of cases with complete data in certain categories, a dichotomous CC status variable (cachectic vs. non-cachectic) was created based on criteria from Fearon et al. (13) wherein cachectic patients reported weight loss >5% over the past 6 months or had a BMI <20 and any weight loss >2% over the past 6 months and non-cachectic cases did not meet either threshold. Quality of life (QOL) was assessed using the European Organization for Research and Treatment of Cancer (EORTC) QLQ-C30 and QLQ-PAN-26 instruments. The QLQ-C30 comprises 30 questions based on QoL, function, and symptoms, and is intended to determine general health-related QoL (HRQoL) specific to cancer patients. With the exception of two items, scoring involves a 4-point Likert scale where 1= “Not at all”, 2= “A little”, 3= “Quite a bit”, and 4= “Very much”. Items 29 (overall health rating) and 30 (overall QoL) are scored on a 7-point ranking ranging from 1= “very poor” to 7= “excellent”. Scales were linearly transformed to a score from 0 to 100 (72) with 100 depicting the best health quality. The EORTC QLQ-PAN26 was validated in PDAC (73–75) and contains 26 items on self-reported symptoms and issues specific to PDAC such as abdominal discomfort and back pain. Higher scores signify poorer QoL for all scales except for the satisfaction scale for which higher scores denote better QoL.

Descriptive statistics were used to summarize baseline characteristics of the study population. For continuous variables, differences between groups were tested by ANOVA with means ± standard deviations (SD) followed by the Tukey Honest Significant Differences post-hoc test. For non-normally distributed variables, non-parametric Kruskal Wallis tests were performed and followed by Mann Whitney U-tests. Chi-square tests or McNemar’s chi square tests were used to determine differences in proportions for categorical variables. Symptoms were also categorized into groups based on the nature of the symptoms. Physical symptoms included six domains (pain, fatigue, drowsiness, nausea, loss of appetite, and shortness of breath). Psychological symptoms included two domains (anxiety and depression). The mean of all symptoms represented the total symptom subscore. Individual symptom scores and subscores obtained from ESAS-r were dichotomized into 0–3 (none to mild) or 4–10 (moderate to severe) (76, 77). The frequency of “none to mild” or “moderate to severe” physical, psychological, and total symptom subscore categories were compared across groups using chi-squared tests.

To identify factors predictive of moderate to severe symptom burden, supportive care needs, HRQoL, and/or CC status, multivariate logistic regression models were used. Pearson correlation tests were used to evaluate the linear relationship between laboratory values and other characteristics. The influence of individual factors and CC status on QoL was investigated with multivariate generalized linear regression. Due to possible collinearity of contributing factors, we performed preliminary global data exploration using multiple correspondence analysis (MCA) (78). MCA is an analog to principal component analysis for categorical variables and was carried out using the R package FactoMiner. A sum of cosine² of >0.3 (a measure of how well each variable is represented by the first 2 dimensions) was used as a cutoff for modeling.

To evaluate associations with OS, we conducted Kaplan-Meier analysis and multivariate Cox proportional hazards (PH) regression, and calculated hazard ratios and 95% confidence intervals (CIs). OS was calculated from the date of diagnosis to the date of death from any cause or last follow-up. Covariates considered in analyses included age at diagnosis, race and ethnicity, gender, BMI, age-adjusted Charlson Comorbidity Index (ACCI), tumor stage, smoking history (never, past, or current), laboratory values, and treatment received (20, 21, 26, 28, 29, 79–82). Effect modification was evaluated by conducting stratified analysis by biological sex and racial and ethnic group. Analyses were performed using R 4.3.1.

A total of 318 PDAC cases (161 female, 157 male) were recruited to the FPC study. Select characteristics of the cohort overall and by race and ethnicity are in Table 2 . The average age at diagnosis was slightly younger in AA (67.2 years) and H/L (66.0 years) compared to NHW (70.5 years), with 25.5% of H/L and 14.3% of AA diagnosed at age 55 or younger compared to 4.2% of NHW (P<0.001 for H/L and P=0.015 for AA pairwise comparisons). A higher proportion of females was observed among AA and H/L racial and ethnic groups compared to NHW, at 62.9%, 66%, and 45.8%, respectively, though significance was only observed for H/L versus NHW pairwise comparisons (P=0.011). NHW were more likely to be married and report a higher annual income than AA and H/L. No significant differences were observed between racial and ethnic groups in education level or insurance status. A higher proportion of never smokers were observed amongst H/L (72.3%) compared to NHW (39.4%), P<0.001. Differences were also observed between racial and ethnic groups for several anthropometric variables. Namely, BMI was slightly higher in AA (28.2 kg/m²) compared to H/L (24.6 kg/m²) and NHW (26.7 kg/m²), with approximately 69% of AA, 60% of NHW, and 42% of H/L reported to be overweight or obese. Waist to hip ratio and waist circumference were significantly lower in H/L compared to NHW, with P=0.0161, P=0.04, respectively (ANOVA and Tukey’s HSD).

Although no statistically significant differences were observed between racial and ethnic groups regarding their personal history of diabetes or the number of comorbidities, a slightly lower proportion of AA participants (34.3%) reported to be fully active/able to carry out pre-disease performance without restriction (ECOG performance status=0) at baseline compared to NHW (41.9%) and H/L (42.6%). Weight loss >5% over the prior 6 months was the most common presenting symptom across all racial and ethnic groups, affecting 62.5% of the cohort. Of cases with known tumor stage documented (n=256; 80.5%), 42.1% had stage I/II disease. When stratified by race and ethnicity, the highest proportion of stage III/IV cases was among AA (51.4%) followed by NHW (36.8%) and H/L (36.2%). Nearly 35% of cases underwent multimodal treatment (defined as receipt of curative-intent surgery and chemotherapy), with receipt of multi-modal treatment highest among H/L (42.6%) and lowest among AA (25.7%). Survival time was significantly longer among H/L (median 22.5 months, 95% CI: 16.6-unestimatable) compared to NHW (14.2 months 95% CI: 12.4–16.7) and AA (12.6 months 95% CI: 10.3–16.3) (P=0.016, log-rank test).

CC status was determined for 309 (97.1%) of the 318 cases with available data. Using criteria from Vigano et al. (15), the prevalence of NCa, PCa, Ca, and RCa at enrollment was 12.9%, 24.6%, 54.1%, and 8.4%, respectively ( Table 2 ; Supplementary Figure 2 ). Thus, approximately 87% of participants were deemed to have pre-cachexia, cachexia, or refractory disease at the time of diagnosis when using the five criteria (15). When using weight loss only to categorize participants as in (13), only 62.5% of individuals were classified as cachectic. More severe stages of CC were observed among males (NCa 10.5%; PCa 17.6%; Ca 62.1%; RCa 9.8%) than females (NCa 15.4%; PCa 31.4%; Ca 46.2%; RCa 7.5%). H/L had the highest proportion of cases with refractory cachexia at diagnosis (17%) followed by AA (11.4%) and NHW (6.2%), though these differences were not statistically significant (P=0.111, univariate chi square test) ( Table 2 ). No associations between baseline BMI and CC stage at baseline were detected (data not shown).

The prevalence of the criterion variables used to categorize CC stage is in Table 3 ; Supplementary Figure 3 . CRP was available in the EMR for only 7% of study participants. We therefore used blood from a subset of participants [n=202; 14.6% AA, 66% NHW, 19.4% H/L)] who donated it for serum processing (46) and evaluated CRP levels using a multiplex panel ( Supplementary Methods ). As expected, the prevalence of abnormal CRP (>10 mg/L) was higher in the RCa cohort (53.8%) compared to 0% of NCa, 25% of PCa, and 16.7% of Ca cases, with statistically significant differences (P<0.001) in CRP levels between CC stages. WBC levels >11,000/mL increased with CC severity, and were observed in 11.8%, 13.17%, and 34.62% of PCa, Ca, or RCa cases, respectively (P=0.0008). Similar to the higher prevalence of abnormal CRP levels among cases classified as PCa compared to Ca, abnormal albumin levels (<3.2 mg/dl) were more common in cases with PCa (11.84%) versus Ca (6.59%). Using a <3.2 mg/dl threshold to identify patients with PCa or Ca appeared to be more sensitive than the lower threshold of albumin (<2.5 mg/dl) used by Vigano et al. (15). Abnormal HgB levels were more prevalent in males than females across the CC continuum, affecting nearly two-thirds of patients classified as having refractory disease. The prevalence of abnormal HgB levels was highest among RCa cases followed by PCa and Ca (P=0.0418 for females, P=0.0007 for males). Laboratory values not considered by Vigano et al. (15) that were significantly different between CC stages using a Cochrane-Armitage trend test included NLR (P=0.0365), GPS (P=0.0059) and bilirubin (P<0.001).

An exploratory evaluation of biomarker levels by race and ethnicity and CC status revealed AA patients were significantly more likely to present with normal CA19–9 levels than other racial and ethnic groups regardless of CC status ( Supplementary Table 1 ), in line with other studies (83). NLR was also more likely to be in the low/normal category among AA, especially among those with CC, with P=0.051. Finally, bilirubin levels were more likely to be abnormal among AA patients with cachexia than patients in the other racial and ethnic groups.

Nearly two-thirds of the 309 cases (n=193; 83 women,110 men) reported loss of >5% of their body weight in the 6 months prior to their diagnosis and were classified as having Ca or RCa ( Tables 2 , 3 ). Decreased food intake (aPG-SGA box 2 score >1) was observed in approximately 15% of cases (n=46) and was most prevalent among those with RCa. Reduced activities and functioning (aPG-SGA box 4 score >2) was reported in 8.1% of cases overall and in nearly 92% of the RCa cases ( Table 3 ). No statistically significant differences were observed by race and ethnicity for weight loss, food intake, or activities/functioning. Of 178 participants who completed questions about hand-grip strength, 148 (83.1%) responded “not at all” to challenges related to hand or finger grip strength, 28 individuals reported having “a little” or “quite a bit” of challenges, and only 2 individuals (with RCa) reported challenges as “very much.” Concordance between self-reported grip strength and dynamometer measurements is unclear.

Symptom burden, which includes symptom type, prevalence, and severity, is summarized overall and by sex and race and ethnicity in Supplementary Table 2 , Supplementary Figure 4 . At baseline, physical and psychological symptom sub-scores were moderate to severe in 43 (14.1%) and 82 (26.9%) of all participants, respectively. The total symptom score was moderate to severe in 19% of cases (n=58). The most commonly-reported physical symptoms across all racial and ethnic groups included fatigue (n=145, 45.5%), loss of appetite (n=108, 35.4%), pain (n=85, 27.9%), and drowsiness (n=68, 22.3%). Of psychological symptoms, anxiety (n=104, n=34.1%) and depression (n=67, 22.0%) were the most frequently reported. While the mean physical subscore was significantly higher among cases with stage III/IV versus stage I/II disease (2.07 versus 1.61, P=0.0354), the median psychological subscore did not differ by stage (P=0.7211) (data not shown). Physical and psychological subscores tended to be higher among females compared to males, though they were not statistically different. More severe anxiety was reported at a higher frequency among females compared to males (P=0.011) and at a lower frequency among AA compared to H/L and NHW (P=0.0346). The prevalence of moderate to severe nausea was significantly higher among H/L and AA participants compared to NHW. Importantly, evaluation of changes in symptom burden over the first 6 months post-diagnosis showed that while mean “tiredness/fatigue” scores nominally increased, “anxiety” and overall psychological scores decreased ( Supplementary Table 3 ).

The problems and supportive care needs reported by participants at baseline are in Supplementary Table 4 , Supplementary Figure 5 . Of 314 participants who completed the Canadian Problem Checklist, emotional concerns ranked first by 177 cases (56.4%) who expressed fears, worries, frustration, and anger. Physical changes involving sleep and weight were the next most frequently cited concerns affecting 46.5% of patients followed by informational concerns (n=130, 41.4%) related to understanding their illness and/or treatment options. No statistically significant differences in concerns were evident when comparing participants by sex or race and ethnicity. Assessment of supportive care needs over the first 6-month time-period revealed significant changes in all major categories (Emotional, Physical, Spiritual, Practical, Informational and Social) when compared to the “None” category ( Supplementary Table 5 ), with significant trends downwards. In contrast, concerns related to concentration/memory, frustration/anger, and worries about changes in appearance increased over time.

HRQoL was evaluated by 180 of 318 PDAC cases. “Tiredness”, “worry”, and “pain” were the main symptoms experienced using the QLQ-30 ( Figure 1A ). Stratification by cachexia status further demonstrated that all functional domains and overall HRQoL significantly worsen as cachexia stage progresses ( Figure 1B ). The distribution of responses to QLQ-PAN26 items plus two items concerning grip strength is shown overall and by cachexia stage in Figures 1C, D . The extent of pain, restrictions in the type and amount of food, and weight loss negatively affected HRQoL, especially among AA and H/L. H/L also reported fewer/less severe side effects from treatment than NHW and AA. Evaluation of responses to the QLQ-PAN26 over time indicates that the “most improved” symptoms are “food taste”, “treatment side effects”, “limited in doing planned activities” and “information from healthcare providers,” each with 56.14% of participants reporting improvements ( Supplementary Table 6 ). Conversely, 42.11% and 47.37% of patients reported a deterioration in HRQoL from “back pain” and “pain at night” respectively. Joy from intimacy, abdominal bloating, and digestive issues worsened for a significant proportion of participants.

MCA (78) of symptom burden, supportive care needs, and HRQoL variables indicated that ~36% of variability was explained by dimensions 1 and 2. ESAS-r categories of pain, appetite, and tiredness/fatigue explained most of the variability along dimension 1, PAN-26 symptom “Gas” explained the majority of variability along dimension 2, and ESAS-r scores for anxiety and depression were explanatory for both dimensions ( Supplementary Figure 6 ). Interestingly and perhaps because physical symptoms may arise from varying clinical sequelae, clustering was only observed in the ESAS-r psychological domain. Binomial regression indicated male sex was negatively associated with an ESAS-r “tiredness” score of >=4 (P=0.03, OR=0.44;CI:0.21,0.90) and an “anxiety” score of >=4 (OR=0.39; 95%CI:0.15,0.95, P=0.045), and positively associated with high bilirubin (P=0.001, OR=2.18; 95%CI:1.06,4.59). Cachexia status was also associated with fatigue, although CIs were wide. Interestingly, patients with a GPS score of >1 were less likely to report “Gas” as a burden (OR=0.158; 95%CI:0.02,0.90, P=0.046). When CC status was modeled via ordinal regression using clinicopathological and demographic predictor variables, we found a significant association between CC status and GPS score >1 (P=0.030; OR=3.67; 95%CI:0.87,12.22). Male sex and late-stage disease also approached significance (P=0.13 and 0.065 respectively). Variance inflation factors were <2 (data not shown).

The median OS of the PDAC cohort was 14.6 months (95% CI: 13.5–16.6 months), with surgical patients having better OS than patients not receiving surgery. Females had slightly longer OS than males, though differences did not reach statistical significance (log-rank P=0.08) ( Figure 2A ). Kaplan Meier curves portrayed that race and ethnicity influenced OS, with H/L having significantly improved survival (HR: 0.57, 95% CI: 0.36–0.89) compared to NHW and AA (Figure 2B). AA had the poorest survival (HR:1.22, 95% 0.79–1.87), but survival time was not statistically different from NHW (P=0.37). Kaplan Meier curves for the four CC stages (15) are in Figure 2C and showed a significant difference (log-rank P=0.0034) between the stages. As expected, those with RCa had the poorest survival (HR: 3.24; 95% CI: 1.63–6.44, log-rank P=0.003) and those with NCa experienced the best survival. No significant differences in survival were observed between the PCa and Ca groups; Ca (HR: 1.87 CI: 1.07–3.28) but not PCa (HR: 1.50 CI: 0.82–2.75) was significantly associated with mortality in univariate analysis. Cachexia status [e.g., cachectic vs non-cachectic using the sole criteria of weight loss over 6 months (13)] also showed significant differences (log-rank test P=0.011) ( Figure 2D ), with cachectic participants having poorer survival on univariate analysis (HR: 1.51, 95% CI: 1.10–2.09). Although a formal analysis of survival for biological sex stratified by race and ethnicity was not performed due to small sample sizes for minorities, we did note an increase in median survival for females for both H/L (female median=34.8 months vs male median=16.6 months) and NHW (female median=16.1 months vs male median=13.5 months) but not AA (female median=12.4 months vs male median=14.9 months) racial and ethnic groups.

Finally, based on findings of univariate analysis we performed PH modeling of OS using cachexia status, biological sex, race and ethnicity (NHW as reference), stage (III/IV vs I/II), GPS (>=2 or <2), NLR (high versus low), and age (early versus late-onset) as predictor variables. Our preliminary model indicated that only cachexia status (HR: 1.82, 95% CI: 1.09–3.0, P=0.021), stage (HR: 2.53, 95% CI: 1.39–4.61, P=0.0002) and GPS (HR:2.48, 95% CI: 1.03–5.96, P=0.01) were significantly associated with survival ( Figure 3A ). When including treatment status (multi-modal versus single therapy), only stage and GPS predicted survival, and upon further interrogation as expected single-therapy treatment status (e.g., palliative/systemic chemotherapy) was highly associated with stage III/IV unresectable disease (P<0.001). In a final more parsimonious PH model, cachectic status, late stage, and a high GPS were all associated with decreased survival times; NLR was not significant in the final model ( Figure 3B ).