{"paper_id":"3580c4e5-b907-4a17-b8fa-b29f7ae7cf44","body_text":"Machine Learning-based Chemotoxicity Predictions in Patients with Colorectal Cancer: Integrating Race, Geospatial Social Determinants of Health, and Biological Aging | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning-based Chemotoxicity Predictions in Patients with Colorectal Cancer: Integrating Race, Geospatial Social Determinants of Health, and Biological Aging Claire Han, Christin Burd, Jesse Plascak, Fode Tounkara, Ashley Rosko, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6628340/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 06 Oct, 2025 Read the published version in BMC Cancer → Version 1 posted 13 You are reading this latest preprint version Abstract Background: Colorectal cancer patients often face chemotoxicity, impacting treatment adherence, survival, and quality of life. Early chemotoxicity screening is vital, yet comprehensive predictive models are lacking. We aimed to develop artificial intelligence(AI)/machine learning (ML)-based models to predict global, gastrointestinal (GI), and hematological chemotoxicity by incorporating racialized group, social determinants of health (SDOH, including Area Deprivation Index measuring geospatial variation) and biological aging (measured by blood-based Levine PhenoAge). Methods: We used electronic health records data from 1,735 adult CRC patients. Sociodemographic/clinical variables, Levine PhenoAge (biological aging), and SDOH (including geospatial data measured by Area Deprivation Index) were analyzed using descriptive statistics. Associations with chemotoxicity (global, GI, hematological) were evaluated via univariate tests. Significant predictors from univariate tests were selected for AI/ML modeling. Six supervised ML models were trained on 80% of cases (n=1,388), with 20% (n=347) reserved for testing. Performance was assessed via accuracy, area under the curve (AUC), and F1-score. Permutation feature importance ranked predictors to define the most significant predictors of chemotoxicity. Results: Chemotoxicity incidences over 6 months of chemotherapy were 56% (global), 41% (GI), and 23% (hematological). Support Vector Machine, followed by XGBoost models (in both training and test datasets) demonstrated high accuracy. Key predictors for global and GI toxicities included advanced biological aging (higher Levine PhenoAge), elevated inflammatory markers (e.g., C-reactive protein), and poor SDOH including geospatial variations (e.g., higher Area Deprivation Index), unemployment. Hematological toxicity was linked to lower immune markers and higher biological age (Levine PhenoAge). Race (non-Hispanic Black), body mass index, and lifestyles also influenced global and GI toxicities. Conclusions : The ML models demonstrated high accuracy in chemotoxicity prediction. Biological aging and SDOH, including ADI, and immune/inflammation markers, were common risk factors of global and GI chemotoxicities. In contrast, biological age and immune/inflammation markers were only linked to hematological chemotoxicity. Integrating these factors into predictive models can help clinicians identify at-risk patients and tailor interventions (e.g., anti-inflammatory and anti-aging strategies) to reduce chemotoxicity and improve survivorship outcomes. Colorectal cancer Chemotoxicity Artificial Intellignece Machine learning Prediction Risk Factors Biological aging Social determinants of health Geospatial variations Figures Figure 1 Figure 2 INTRODUCTION Colorectal cancer (CRC) remains a significant public health burden in the United States, with an estimated 153,020 new cases and 52,550 deaths projected for 2024 (1). Despite advances in screening and treatment, CRC is the third most commonly diagnosed cancer and the second leading cause of cancer-related mortality in the U.S. (1). 5-Fluorouracil (5-FU)-based chemotherapy is a cornerstone of CRC treatment, particularly for patients with stage II-III CRC (2). While 5-FU-based regimens have improved survival outcomes, they are associated with significant challenges, including severe chemotoxicity that can compromise treatment efficacy and patient quality of life in CRC (2-6). These side effects underscore the need for a deeper understanding of the factors influencing chemotoxicity and its management. Chemotherapy-induced toxicity is a major concern in CRC treatment, with up to 30-40% of patients experiencing severe adverse effects such as hematological toxicity (e.g., neutropenia, leucopenia), gastrointestinal (GI) toxicity (e.g., colitis, diarrhea, abdominal pain, gastrointestinal bleeding), neurotoxicity, and cardiotoxicity (7-10). Of note, GI and hematological complications are the most frequent and severe in CRC (7-10). These complications not only cause physical distress but also contribute to psychological burdens, including anxiety, depression, and worsened pre-existing stress (6). The prevalence of chemotherapy-related complications is alarmingly high, requiring dose reductions, treatment delays, or discontinuation due to toxicity (8, 11, 12). Such interruptions in therapy are associated with poorer survival outcomes, increased hospital admissions, and frequent emergency department visits, further straining healthcare resources (13). Additionally, chemotoxicity significantly impairs patients' quality of life, affecting their ability to perform daily activities and maintain social relationships (14, 15). These challenges highlight the critical need for effective strategies to predict, prevent, and manage chemotherapy-related toxicity in CRC patients. Despite the widespread use of 5-FU-based chemotherapy, significant knowledge gaps persist in understanding and managing chemotoxicity. Previous studies have been limited by small sample sizes, lack of racial and ethnic diversity, and insufficient attention to social determinants of health (SDOH) that may influence toxicity risk in CRC (7-10, 16). Most research in chemotoxicity prediction in cancers has focused on non-Hispanic White populations with other cancer types (primarily breast and lung cancers), leaving underrepresented groups, such as African American/Black, Hispanic or Latino, Asian, and Indigenous patients, inadequately studied (17-19). Emerging evidence suggests that racialized groups and SDOH play a critical role in cancer health outcomes, such as chemotoxicity (20, 21). Studies have shown that racial and ethnic minoritized groups often experience higher rates of chemotherapy-related complications, which may be attributed to disparities in access to care, socioeconomic status, or other social factors, and underlying comorbidities (20, 21). Additionally, chronic stress, which is more prevalent in marginalized populations, has been linked to dysregulated immune function and increased susceptibility to chemotoxicity(22-24). For example, Black and Hispanic CRC patients are more likely to report severe symptom burden and impaired quality of life during chemotherapy compared to White patients (16). These disparities highlight the importance of incorporating race and SDOH into toxicity risk prediction models. Residential socioeconomic conditions should also be included and can be measured through validated measures relying on socioeconomic surveys of residents administered by the US Census Bureau (25, 26). Therefore, our study suggested integrating geospatial SDOH measured by the Area Deprivation Index (ADI) to understand the CRC population's chemotoxicity risk. Furthermore, emerging evidence shows that baseline biological aging is associated with immune functions, which predict chemotoxicity in cancer patients (27-29). Levine PhenoAge is a validated biological aging marker that can be computed using routine circulatory blood samples, including immune and inflammation markers, complete blood cell counts, and liver and kidney functions, without further blood assays (30). Therefore, baseline Levine PhenoAge may have the potential as a biological aging marker in predicting chemotoxicity, requiring further validation in CRC. Given the limited understanding of risk factors of chemotoxicity mentioned above, there is a lack of reliable risk prediction tools, including comprehensive data to identify patients at high risk for chemotoxicity in CRC, which hinders proactive monitoring and personalized management strategies (8, 11). The current \"one-size-fits-all\" approach to chemotherapy management has proven inadequate, as it fails to account for the heterogeneity in toxicity profiles among patients. For instance, while some patients experience severe GI toxicity, others may develop hematological toxicity or other toxicities (e.g., neurotoxicity or cardiotoxicity), each requiring distinct management approaches (8, 11). This underscores the need for a more in-depth understanding of the types and mechanisms of chemotoxicity across racially and socially diverse and larger samples, to develop targeted interventions. To address these gaps, our study leveraged the electronic health records (EHRs) including SDOH (e.g., geographic variations), from a large and racially diverse cohort of CRC patients. In our study, we applied artificial intelligence (AI)/machine learning (ML) methods that offer advantages over traditional statistical methods, such as multivariate regression, to develop predictive models for chemotoxicity in CRC patients. Unlike regression models, which assume linear relationships and struggle with high-dimensional data, AI/ML can capture complex, non-linear interactions among clinical, biological, and social factors (31-34). AI/ML also handles large, diverse datasets, such as EHRs, which include structured data (e.g., lab results) and unstructured data (e.g., clinical notes). Moreover, AI/ML outperforms traditional methods in predictive performance compared to regression-based approaches (31-34). Therefore, the aim of this study was to 1) develop AI/ML-based chemotoxicity prediction models (global, GI, and hematological chemotoxicity incidences) over the 6 months of 5-FU-based chemotherapy, and 2) identify the importance of chemotoxicity risk factors at baseline, including race, biological aging markers (i.e, Levine PhenoAge), and SDOH (e.g., geospatial variations measured by ADI) . We hypothesized that our AI/ML models would demonstrate high accuracy of chemotoxicity prediction, baseline biological aging markers (Levine PhenoAge), ADI, and race would be significant predictors of chemotoxicity risk, and key risk factors may differ by types of chemotoxicity. By leveraging AI/ML, our study aims to develop more accurate and inclusive prediction models for chemotoxicity. This will enable proactive monitoring and personalized management, ultimately improving treatment outcomes and quality of life, and reducing disparities in CRC. METHODS Study Design, Setting, and Data Sources This retrospective cohort study utilized EHR data from the Ohio State University (OSU) Comprehensive Cancer Center, covering the data from January 1, 2010, to December 31, 2020. The Ohio State University Comprehensive Cancer Center is a large not-for-profit academic cancer center in the Columbus, Ohio region with a catchment area encompassing the entire state. The OSU Honest Broker Operations Committee (HBOC) service has identified the eligible cohort and provided de-identified EHR data for our study in compliance with the US Health Insurance Portability and Accountability Act (HIPAA) (35). The study population included 1,735 adult patients diagnosed with stage II-III CRC who received 5-FU-based chemotherapy infusions. Inclusion Criteria include: Adult patients aged > 18 with stage II-III CRC as the primary cancer site; underwent colon or rectal surgery without a stoma; scheduled for 8-12 cycles of 5-FU-based chemotherapy infusions (5-FU alone, Xeloda, or FOLFOX); single primary cancer; and having necessary baseline data including sociodemographic and clinical data, routine blood tests at baseline, and 6 months post-chemotherapy initiation, and zip codes were included. Exclusion Criteria include : Patients having a current ostomy, chronic bowel disorders; regular use of steroids/immune suppressants; history of neoadjuvant chemotherapy; on active radiation or immune therapies as primary cancer therapies; and pregnant women; had incomplete or missing EHR data required for this study at baseline before chemotherapy and 6 months after chemotherapy were excluded. Measures D emographic and Clinical Dat a Demographic data include race, ethnicity, chronological age, sex, and marital status. Clinical data include chemotherapy regimens, doses, and durations/cycles, weight, diet and exercise habits, smoking status, alcohol consumption, cancer stages, time since CRC diagnosis, history of radiotherapy, immunotherapy or surgery, Charlson Comorbidity Index, and blood-based immune and nutritional markers (e.g., albumin, platelets, hemoglobin). Among these variables, we controlled sex, cancer sites, and insurance types as covariates. Biological Aging Markers (Levine PhenoAge) We used Levine PhenoAge to measure biological aging. In our study, the Levine PhenoAge was determined from routine monthly blood tests and included nine biomarkers in accordance with the method described previously. This method, validated in multiple studies including cancer patients, is based on the following measures (30, 36): Albumin (g/L) and Alkaline Phosphatase (U/L) —Liver function, Creatinine (umol/L)—Kidney function, Glucose (mmol/L)—Metabolism, C-reactive protein (CRP) (mg/dl)—Inflammation, Lymphocyte (%), Mean Cell Volume (MCV) (Fl), Red Cell Distribution Width (RCDW) (%), White Blood Cell Counts (WBC) (1K/ul)—Immune system, and Chronological age (years). SDOH, including geospatial variations (ADI) We included SDOH-related variables from the EHR: employment status and geospatial variations measured by ADI (i.e., residential zip code-based socioeconomic and environmental deprivation). Other SDOH variables—like education levels, home ownership, annual household income, and poverty levels—were excluded due to a significant amount of missing data in the EHR. ADI: The ADI measures Residential zip code-based socioeconomic and environmental deprivation (ADI) . The ADI is a composite measure scored from 1 (least disadvantaged) to 100 (most disadvantaged), calculated using weighted coefficients from 17 indicators, such as income, education, employment, population age, poverty, and housing conditions (37). Developed by the University of Wisconsin, the ADI dataset is structured by ZIP codes +4 geographic areas (38). For this study, the 2015 ADI version was used to match patient ZIP codes +4 geographic areas available in the EHR, and ADI scores were divided into tertiles, with the highest tertile indicating the most socioeconomically disadvantaged group, following prior validation methods (37-39). Outcome s : Chemotoxicity i ncidence s over 6 months of 5-FU-based chemotherapy We assessed global, GI, and hematological chemotoxicity incidences that occurred from baseline to 6 months post-chemotherapy initiation. Chemotoxicity incidences were measured as dichotomized outcomes (presence versus absence) over the 6 months of chemotherapy using the clinician-reported adverse events. The chemotoxicity was reported by clinicians, based on laboratory or diagnostic data using the Common Terminology Criteria for Adverse Events criteria (CTCAE v.5.0: Grade I ‘mild’, II ‘moderate’, III ‘severe’, IV ‘life-threatening’, and V ‘death related to adverse events’) (reliability=0.85, sensitivity=0.79) (40). The OSU HBOC service used the International Classification of Diseases (ICD) codes 9 or 10 to identify chemotoxicity. Data Analyses Overall Statistical Methods Descriptive statistics were used to summarize the sociodemographic and clinical characteristics of the study population. A p-value of < 0.05 was applied to determine statistical significance. We used R software, the MLR (Machine Learning in R) R package (version 3.6.3, R Foundation for Statistical Computing, Vienna, Austria), and Python (version 3.10.2, Python Software Foundation, Wilmington, U.S.) when appropriate for statistical analyses. Then, we evaluated associations of potential predictors (sociodemographic and clinical variables, Levine PhenoAge, and SDOH including geospatial variations) with chemotoxicity using the Chi-square test, or analysis of variance (ANOVA) for chemotoxicity incidences (as categorical variables) in whole datasets. We included three types of chemotoxicity incidences (global, GI, and hematological) as outcome variables, and performed all analyses separately. Given the purpose of our study to identify baseline predictors of chemotoxicity, we did not adjust for post-baseline measures of time-varying variables (i.e., blood-based immune markers and Levine PhenoAge). This step identified significant factors for inclusion in the AI/ML model training. Potential predictors were included only if they were significant in initial analyses. AI/ML Approach AI/ML Modeling. Among the 1,735 adult CRC patients, all had Levine PhenoAge data, and ZIP codes, and most of the features (i.e., risk factors) used were available with minimal missing data (missing data rates < 5%). Given the low rates of missing data, we conducted complete case analyses. Outliers were identified and removed using the interquartile range (IQR), and duplicate entries were eliminated to prevent model skewing. Then, we performed data transformation, including normalization (scaling numerical features to a range of 0–1) to ensure equal contribution of all features to the models, and standardization (adjusting to a mean of 0 and a standard deviation of 1) to reduce feature dimensions and improve the performance and stability of our ML models [49]. Ordinal encoding was applied to categorical features with a natural order. AI/ML Model Classification and Evaluations. We first constructed a training dataset by selecting 80% of the CRC patients (n = 1,388) and reserving the remaining 20% (n = 347) for the test dataset. We matched age groups (10-year intervals), race, sex, and incidences of chemotoxicity (global, GI, and hematological toxicities) between training and test datasets. Using Python’s `train_test_split` function from the scikit-learn (sklearn) library, samples were assigned to either the training or testing set (41). We also compared the comparability between training and test datasets (Table 1). Covariates were controlled for AI/ML modeling. The training dataset was used for initial model development, and the test dataset was employed to evaluate the model’s performance in predicting chemotoxicity incidences. We examined the data compatibility between training and test datasets using Chi-square tests or ANOVA (Table 1). We applied six supervised ML methods: logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), gradient boosting machine (GBM), and extreme gradient boosting (XGBoost) (42). Each method was evaluated using a specific learning algorithm to assess its effectiveness in chemotoxicity predictions. Logistic regression, a widely used binary classifier, served as the baseline model for comparison. To prevent overfitting, hyperparameter tuning was performed using a random search with 5-fold cross-validation. The 5-fold cross-validation results were evaluated using the mean value of each model performance metric with a 95% confidence interval (CI) (42, 43). Model performance was assessed using metrics such as accuracy, precision, recall (sensitivity), specificity, F1 score, and area under the receiver operating characteristic curve (AUC) (42, 43). Permutation Feature Importance (PFI) . To identify and rank the most significant risk factors for predicting chemotoxicity, a feature importance analysis was performed (44). Using PFI, the contribution of each feature was quantified by assessing its impact on the performance of the best-performing ML models. This involved randomly shuffling each feature 10 times individually within the trained classification model and measuring the resulting decline in model accuracy. The analysis utilized hold-out test data to evaluate the influence of features on chemotoxicity outcomes. RESULTS Initial Descriptive Analyses Table 1 outlines the demographic and clinical characteristics, biological aging markers (Levine PhenoAge), SDOH including geospatial variations (ADI), and chemotoxicity outcomes for the training (n = 1,388, 80% of the total sample of 1,735 adults with CRC) and test (n = 347, 20% of the total sample) datasets. No significant differences were observed between datasets. The mean age of participants was 65.3 ± 13.2 years in the training dataset and 64.2 ± 12.6 years in the test dataset (p = 0.352). Levine PhenoAge was similar between the training (70.3) and test (70.1) datasets (p = 0.423). Approximately half of the participants were women in both datasets. The majority of patients were non-Hispanic White (67.2% in the training dataset and 72.0% in the test dataset), followed by non-Hispanic Black in both datasets (p = 0.984). We did not have Hispanic populations in our datasets. Most participants were married, had colon cancer, and were in Stage II, and had a high body mass index (BMI) ‘overweight or obese’, higher WBC and CRP levels than normal ranges, and a history of GI surgery. The neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) (higher levels indicating a pro-inflammatory status), were also similar, with the majority of patients within the normal range across the datasets. Most were married, insured (mainly private), employed, and had Stage II colon cancer. Chemotoxicity incidences were similar across datasets: global (56.3%), GI (40.7%), and hematological (22.8%) in total samples. Potential Risk Factors for Chemotoxicity Table 2 summarizes and highlights potential risk factors significantly associated with chemotoxicity incidences, categorized into global, GI, and hematological chemotoxicity using an ANOVA or a Chi-square test. Chronological age was not associated with any type of chemotoxicity. Biological age was higher than chronological age in our datasets. Higher biological age (Levine PhenoAge) was a consistent risk factor across all chemotoxicity types. Non-Hispanic Black individuals were at higher risk for global and GI chemotoxicity, while non-Hispanic White individuals were at lower risk. Married/partnered marital status was negatively associated with chemotoxicity risk (Global and GI). Greater comorbidities were associated with global and GI chemotoxicity risk. Both underweight (BMI mean 17, range 15-22) and obesity (BMI > 30) were risk factors for GI and global chemotoxicity, respectively. Elevated WBC, CRP, and NLR were strongly associated with increased risk for global and GI chemotoxicities, while lower WBC and NLR were associated with hematological toxicity risk. Previous cancer treatment histories, including radiation, immunotherapy, and GI surgery, were associated with higher global and GI chemotoxicities. In terms of health behaviors, heavy alcohol consumption increased GI chemotoxicity risk, while regular physical activity reduced global chemotoxicity risk. Higher ADI was linked to increased global and GI chemotoxicities, and unemployed or retired status influenced GI chemotoxicity risk. Detailed results are described in Supplementary Tables 1-3. Performance Comparison for Classification Models We evaluated the performance of six ML models using a training dataset for development and a test dataset for validation, with five-fold cross-validation for global chemotoxicity, GI chemotoxicity, and hematological chemotoxicity (Table 3). For AI/ML chemotoxicity prediction modeling, we included potential risk factors showing significant associations with chemotoxicity in prior analyses (Table 2). For global chemotoxicity in the training dataset, SVM showed the highest AUC (0.988), accuracy (0.962), precision (0.954), sensitivity (0.971), specificity (0.953), and F1 score (0.963). In the test dataset, SVM also demonstrated the best performance with the highest AUC (0.987), accuracy (0.960), precision (0.949), sensitivity (0.971), specificity (0.948), and F1 score (0.960), indicating the superior ability to distinguish between positive and negative cases. For GI chemotoxicity in both training and test datasets, SVM achieved the highest AUC (0.984) in both datasets, as well as the highest values in other evaluation metrics, followed by XGBoost model, with overall good performance metrics shown in Table 3. For hematological chemotoxicity, similar results are shown; the SVM model achieved exceptional performance across all metrics in both training and test datasets, followed by the XGBoost model. The AI/ML prediction model AUC-ROC (Figure 1) also confirmed our findings in Table 3. Feature Importance The most significant features (based on PFI analyses) of the best-performing model (SVM in our study) in the training and test datasets were analyzed for each type of chemotoxicity. Among the significant input features identified in Table 2, the most important features for each type of chemotoxicity were overall consistent between the training and test datasets, except for employment status and previous history of cancer treatments. Figure 2 illustrates the relative importance of the input features of the AI/ML models. Global Chemotoxicity . The SVM models in the training and the test dataset were evaluated separately using permutation feature importance (Figure 2A: orange bars represent the training dataset, and blue bars represent the test dataset). Significant features predicting global chemotoxicity in both datasets included being divorced/widowed or single, lack of physical activity, higher CRP levels, increased biological age (Levine PhenoAge), and higher ADI. Notably, being divorced/widowed or single and lack of physical activity were the most important features in the training dataset, while high CRP and being divorced/widowed or single were the most significant in the test dataset. Levine PhenoAge and ADI were consistently important features for global chemotoxicity in both datasets. Other contributing factors with minor impacts included greater comorbidities, higher NLR, racial minority status (primarily non-Hispanic Black), and obesity. In the training dataset, a previous history of cancer treatments was identified as a significant feature, while in the test dataset, being unemployed or retired was also significant, though both features had only a minor impact on global chemotoxicity. GI Chemotoxicity . The SVM models in both training and test datasets were evaluated separately (Figure 2B). Significant features predicting GI chemotoxicity in both datasets included unemployed or retired status, high CRP levels, being divorced or single, higher biological age (Levine PhenoAge), higher ADI, higher WBC count, heavy alcohol consumption, higher NLR, racial/ethnic minority status (primarily non-Hispanic Black), greater comorbidities, lack of physical activity, and lower-than-normal BMI. Hematological Chemotoxicity . The SVM models in both training and test datasets were evaluated separately (Figure 2C). Significant features predicting hematological chemotoxicity in both datasets included lower WBC count, lower NLR, and higher biological age (Levine PhenoAge). Levine PhenoAge was a predictor of hematological toxicity, while ADI was not. Other SDOH variables did not contribute to predicting hematological toxicity. DISCUSSION This research is the first time to 1) develop AI/ML chemotoxicity prediction models including various common types of chemotoxicity (Global, GI, and hematological toxicities); 2) integrate a wide range of factors including biological aging and SDOH including geospatial variations; and 3) identify the most significant risk factors of chemotoxicity in patients with CRC. Additionally, our model used longitudinal EHR data, which improved chemotoxicity prediction. Our study demonstrated that the AI/ML algorithms successfully built prediction models with high accuracy, ranging from good (0.5 ≤ AUC < 0.7) to moderate to high (>0.7)(33, 41, 42). The study also highlighted the relative importance of specific features in predicting chemotoxicity, revealing that biological age, race, and certain SDOH factors, including ADI, play a significant role in predicting chemotoxicity risk. The ML models developed and validated in this study have the potential to guide personalized strategies for identifying CRC patients at high risk of chemotoxicity, enabling tailored interventions to address unmet needs contributing to chemotoxicity risk in this population. In our univariate analyses (ANOVA), chronological age was not associated with any types of chemotoxicities (Table 2). Conversely, biological age, as measured by Levine PhenoAge, was positively correlated with increased chemotoxicity across all categories (global, GI, and hematological) in univariate analyses as well as in AI/ML models. Biological age (measured using Levine PhenoAge) was higher than chronological age in our samples. Our study supports that biological aging, rather than chronological age alone, may be a more robust predictor of chemotoxicity risk. SDOH were significantly associated with chemotoxicity risk in our study. Biological aging, rather than chronological age, was a stronger predictor of chemotoxicity. SDOH, including higher ADI scores, unemployment or retirement, and being unmarried, were linked to greater global and GI toxicities, underscoring the impact of socioeconomic and social support factors. Non-Hispanic Black individuals experienced higher toxicity rates, likely reflecting broader structural racial inequities, including access to healthcare, socioeconomic position, and neighborhood disadvantage (45-50). However, our study sample lacked representation of other racial and ethnic minority groups, such as Hispanic populations, underscoring the need for future research in more diverse cohorts to better understand the complex relationships between SDOH and cancer treatment outcomes (51). Our findings added new knowledge to the existing literature (52, 53) by providing evidence that pro-inflammatory status (high WBC, CRP, and NLR) (52) and greater comorbidity (53) are associated with cancer health outcomes, including chemotoxicity (specifically global and GI toxicity as well). Our findings highlight the role of pre-existing health conditions in exacerbating chemotoxicity. These findings suggest that systemic inflammation may serve as a key mediator of chemotoxicity, potentially exacerbating tissue damage and impairing recovery (52). Interventions targeting inflammation, such as anti-inflammatory therapies or lifestyle modifications, may help mitigate these risks. In our study, BMI exhibited a complex relationship with chemotoxicity. While higher BMI (>30 kg/m²) was associated with increased global toxicity, lower BMI (<25 kg/m²) was linked to higher GI toxicity, particularly in underweight individuals with a mean BMI of 17 (range: 15–22). This underscores the potential dual role of BMI as both a protective and a risk factor of chemotoxicity. The complex relationship between high BMI and chemotoxicity can be explained as follows. For higher BMI and increased global toxicity, obesity may alter drug pharmacokinetics, increase systemic inflammation, and exacerbate comorbidities, leading to higher global toxicity (54, 55). Obese patients are also often prescribed large doses of chemotherapy, leading to excessive toxic effects (56). Low BMI, especially in underweight individuals, reflects poor nutritional status, reduced physiological reserve, and cachexia, making the GI tract more vulnerable to chemotherapy damage (57). Personalized chemotherapy dosing and supportive care (e.g., nutritional support for underweight patients and inflammation management for obese patients) are needed to mitigate toxicity risks with BMI. Previous history of cancer treatments, including radiation, immunotherapy, and GI tract surgery, were associated with increased global and GI toxicity. This suggests that cumulative treatment burden may heighten susceptibility to chemotoxicity, particularly in patients with a history of aggressive or multimodal therapies. These findings highlight the importance of considering treatment history when assessing chemotoxicity risk and designing personalized treatment plans. Lifestyle behaviors, such as heavy alcohol use, were associated with increased GI toxicity, while regular physical activity was linked to reduced global toxicity. These findings underscore the potential of lifestyle interventions, such as alcohol reduction and exercise programs, in mitigating chemotoxicity risk (58). Integrating behavioral support into cancer care may improve treatment tolerance and long-term outcomes (58). We found that greater biological age and lower WBC and NLR levels were significant risk factors for hematological toxicity, while clinical and SDOH factors were not. Similar findings were observed in other studies (8, 59, 60). This suggests hematological toxicity is primarily driven by intrinsic aging and immune/inflammation-related biological factors, such as reduced bone marrow reserve and immune capacity, rather than external clinical or socioeconomic influences (8, 59-61). Notably, the association between low NLR and toxicity should be interpreted with caution, as genetic variants like the Duffy-null phenotype—common in individuals of African descent—can result in lower ANC despite normal immune function, potentially influencing NLR values (62). Clinical Implications. The identification of these risk factors has important clinical implications. AI/ML models incorporating biosocial markers could help identify CRC patients at high risk for chemotoxicity, enabling early interventions and personalized treatment strategies. For instance, patients with elevated inflammatory markers or high ADI scores may benefit from closer monitoring and targeted supportive care. Additionally, addressing modifiable risk factors, such as lifestyle behaviors and comorbidities, including obesity or malnutritional status, may improve treatment tolerance and quality of life. Lastly, oncologists should screen high-risk groups for different types of chemotoxicity (e.g., GI vs. hematological) and provide tailored interventions specific to each chemotoxicity type. Strengths and Limitations. Our study leverages diverse input data, including biological aging and immune markers, and is not limited to non-Hispanic White samples. It incorporates longitudinal data, objective SDOH measures (e.g., ADI), and a large EHR dataset, enhancing the rigor of AI/ML modeling. However, as a secondary analysis of EHR data, our study may lack key variables, such as representation of other racial/ethnic minorities (e.g., Hispanic populations) and additional SDOH factors (e.g., income, home ownership, poverty levels, occupation, healthcare access, and social support). Additionally, while we controlled for changes in blood markers including Levine PhenoAge over time in our analyses, we did not examine their predictive role in chemotoxicity. Future longitudinal studies with multiple time points are needed to better understand the longitudinal predictability of changes in biological aging over time for chemotoxicity. Future research should focus on expanding racial/ethnic diversity, incorporating comprehensive SDOH measures, and exploring longitudinal dynamics of time-varying factors including immune and biological aging markers, to refine chemotoxicity prediction models. External validation of our study using different datasets is needed to confirm our AI/ML models. Additionally, further research to investigate underlying biosocial mechanisms of chemotoxicity and tailored intervention strategies, is suggested. CONCLUSION The AI/ML chemotoxicity prediction models demonstrated high accuracy using SVM and XGBoost algorithms. This study highlights key predictors of chemotoxicity, including marital status (divorced/widowed or single), higher biological age (Levine PhenoAge), increased immune and inflammatory markers (e.g., WBC, CRP, NLR), and SDOH disparities such as geospatial variations (higher ADI) and employment status (unemployed or retired). Non-Hispanic Black race and BMI also play roles, with higher BMI associated with global toxicities and lower BMI with GI toxicities. Lower WBC, NLR, and higher biological age (Levine PhenoAge) predict greater hematological chemotoxicity. Risk factors vary by type of chemotoxicity (global, GI, and hematological). Findings emphasize the need for personalized treatment, addressing disparities, and targeted interventions (e.g., reducing inflammation, anti-aging strategies, and improving health disparities). These findings underscore the importance of a holistic approach to cancer care that addresses biological and social health determinants. By integrating these factors into predictive models, clinicians can better identify at-risk CRC patients and implement tailored interventions to mitigate chemotoxicity and improve survivorship outcomes. Abbreviations Abbreviation Full Term 5-FU 5-Fluorouracil ADI Area Deprivation Index AI Artificial Intelligence AUC Area Under the Curve BMI Body Mass Index CI Confidence Interval CRC Colorectal Cancer CRP C-Reactive Protein CTCAE Common Terminology Criteria for Adverse Events DT Decision Tree EHR Electronic Health Records F1 F1 Score (harmonic mean of precision and recall) GBM Gradient Boosting Machine GI Gastrointestinal ICD International Classification of Diseases IQR Interquartile Range LR Logistic Regression ML Machine Learning NLR Neutrophil-to-Lymphocyte Ratio OSU Ohio State University PFI Permutation Feature Importance PLR Platelet-to-Lymphocyte Ratio R Statistical software R RCDW Red Cell Distribution Width RF Random Forest SDOH Social Determinants of Health SVM Support Vector Machine WBC White Blood Cell count XGBoost Extreme Gradient Boosting Declarations This study is a secondary data analysis using de-identified EHR data, Thus, the IRB approval was waived for the current study. IRB Ohio State University Supplementary Materials: The following supporting information (Supplementary Tables 1-3) can be downloaded at the journal site. Author Contributions: Conceptualization, CH, XN, CB, JP, AN, DV.; methodology, CH, XN, TF, AT; validation, CH, XN, CB, JP, FT, AT, DV; formal analysis, CH, FT.; investigation, CH, XB, CB, JP, AR, AT, DV; data curation, CH, FT.; writing—original draft preparation, CH; writing—review and editing, CH, XN, CB, JP, AN, AR, AT, DV; visualization, CH, FT.; supervision, XN. All authors have read and agreed to the published version of the manuscript. Funding: The work was original research that had not been published previously. CH : Cancer Research Seed Grant from the Ohio State University College of Nursing and the Ohio State University Comprehensive Cancer Center. CH is also funded by the Oncology Nurse Foundation (ONF) RE03. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Institutional Review Board Statement: Thank you for your feedback. Our study utilized de-identified secondary electronic health record (EHR) data from the Ohio State University Comprehensive Cancer Research Center. As this research involved retrospective analysis of pre-existing, anonymized data, institutional review board (IRB) approval was not required. The Ohio State University Cancer Center’s IRB classified this study as non-human subjects research and granted an exemption per institutional policy and U.S. regulations (45 CFR 46). A waiver of informed consent was also granted. All data were fully anonymized and handled in compliance with HIPAA’s Privacy Rule. Access to the dataset was governed by Ohio State University Comprehensive Cancer Research Center’s data governance policies and a formal data-use agreement. Informed Consent Statement: Not applicable. Data Availability Statement: The datasets analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request. Our study did not generate any data and we used the de-identified EHR data as a secondary data analysis. Furthermore, the de-identified dataset used in this study contains restricted geographic (zip code) and demographic (age >80) elements that may pose re-identification risks under HIPAA. To comply with institutional and regulatory requirements, access is limited to qualified researchers through a formal data request process overseen by OSU, subject to a Data Use Agreement and institutional review. Acknowledgments: Dr. Andrey Loginov, PhD, Bioinformatician, AI/ML Engineer, and Big Data Scientist, University of Maryland School of Medicine, Baltimore, Maryland, United States. We thank Dr. Loginov for his assistance in data curation and AI/ML modeling. Conflicts of Interest: The authors declare no conflicts of interest. References Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin . 2024;74(1). Doi:10.3322/caac.21820 Chan GHJ, Chee CE. Making sense of adjuvant chemotherapy in colorectal cancer. J Gastrointest Oncol . 2019;10(6):1183-92. Doi:10.21037/jgo.2019.09.03 Biller LH, Schrag D. Diagnosis and treatment of metastatic colorectal cancer: a review. JAMA . 2021;325(7):669-85. Doi:10.1001/jama.2021.0106 Extermann M. Chemotherapy toxicity. In: Gu D, Dupre ME, editors. Encyclopedia of gerontology and population aging . Cham: Springer; 2020. P. 1-6. Doi:10.1007/978-3-319-69892-2_1001-1 Sargent DJ, Goldberg RM, Jacobson SD, et al. A pooled analysis of adjuvant chemotherapy for resected colon cancer in elderly patients. N Engl J Med . 2001;345(15):1091-7. Doi:10.1056/NEJMoa010957 Tantoy IY, Cataldo JK, Aouizerat BE, et al. A review of the literature on multiple co-occurring symptoms in patients with colorectal cancer. Cancer Nurs . 2016;39(6):437-45. Doi:10.1097/NCC.0000000000000336 Han C, Yang G, Syrjala K. Symptom experiences in colorectal cancer survivors: a systematic review and meta-analysis. Cancer Nurs . 2020; 43(3):E132-E158. Doi:10.1097/NCC.0000000000000678 Han CJ, Ning X, Burd CE, et al. Chemotoxicity and associated risk factors in colorectal cancer: a systematic review. Cancers (Basel) . 2024;16(14). Doi:10.3390/cancers16142662 Han CJ, Reding KW, Kalady MF, et al. Factors associated with long-term gastrointestinal symptoms in colorectal cancer survivors. PloS One . 2023;18(5):e0286058. Doi:10.1371/journal.pone.0286058 Gustavsson B, Carlsson G, Machover D, et al. A review of systemic chemotherapy in colorectal cancer. Clin Colorectal Cancer . 2015;14(1):1-10. Doi:10.1016/j.clcc.2014.11.002 Feliu J, Espinosa E, Basterretxea L, et al. Prediction of chemotoxicity in older patients with colorectal cancer. Cancers (Basel) . 2022;14(1):127. Doi:10.3390/cancers14010127 Yazbeck V, Alesi E, Myers J, et al. An overview of chemotoxicity and radiation toxicity in cancer therapy. Adv Cancer Res . 2022;155:1-27. Doi:10.1016/bs.acr.2022.02.001 Lyman GH, Yau L, Nakov R, et al. Overall survival and risk of second malignancies with chemotherapy and G-CSF. Ann Oncol . 2018;29(9):1903-10. Doi:10.1093/annonc/mdy244 Andreyev J, Adams R, Bornschein J, et al. British Society of Gastroenterology guidance on gastrointestinal complications of cancer treatment. Gut . 2025. [Epub ahead of print]. Doi:10.1136/gutjnl-2024-332115 Baltussen JC, de Glas NA, van Holstein Y, et al. Chemotherapy-related toxic effects and quality of life in older patients. JAMA Netw Open . 2023;6(10):e2339116. Doi:10.1001/jamanetworkopen.2023.39116 Han CJ, Tounkara F, Kalady MF, et al. Racial/ethnic disparities in HRQoL in colorectal cancer survivors. J Gastrointest Cancer . 2024;55(3):1179-89. Doi:10.1007/s12029-024-01024-8 Alibhai SM, Breunis H, Gregg RW, et al. Validating the CARG toxicity tool in prostate cancer. J Clin Oncol . 2019;37(15_suppl):e23081. Doi:10.1200/JCO.2019.37.15_suppl.e23081 Moth EB, Kiely BE, Stefanic N, et al. Predicting chemotherapy toxicity in older adults. J Geriatr Oncol . 2019;10(2):202-9. Doi:10.1016/j.jgo.2018.08.004 Pathak N, Nishijima TF, Cavdar E. The applicability of the CARG chemotherapy toxicity tool. J Geriatr Oncol . 2024;15(8):102070. Doi:10.1016/j.jgo.2023.102070 Nipp RD, Shui AM, Perez GK, et al. Health care access among cancer survivors. JAMA Oncol . 2018;4(6):791-7. Doi:10.1001/jamaoncol.2018.0097 Lorentsen MK, Sanoff HK. Social determinants of health and colorectal cancer outcomes. Curr Treat Options Oncol . 2024;25(4):453-64. Doi:10.1007/s11864-024-01194-4 Harris AR, Pichardo CM, Franklin J, et al. Multilevel stressors and tumor immunity in breast cancer. JAMA Netw Open . 2025;8(2):e2459754. Doi:10.1001/jamanetworkopen.2024.59754 Goel N, Hernandez A, Cole SW. Social genomic determinants of health. J Clin Oncol . 2024;42(30):3618-27. Doi:10.1200/JCO.23.02145 Tran T, Rousseau MA, Farris DP, et al. Social vulnerability index in cancer research. Cancer Causes Control . 2023;34(5):407-20. Doi:10.1007/s10552-023-01677-z Pearson J, Jacobson C, Ugochukwu N, et al. Geospatial analysis of social determinants of health. Int Anesthesiol Clin . 2023;61(1):49-62. Doi:10.1097/AIA.0000000000000389 US Census Bureau. American Community Survey, 2009-2013. [Internet]. Available from:https://data.census.gov/. Accessed March 17, 2025. Fell G, Rosko AE, Abel GA, et al. Peripheral blood CD3+ T-cell biomarkers in frailty. Br J Haematol . 2022;199(1):100-5. Doi:10.1111/bjh.18340 Rosko AE, Elsaid MI, Woyach J, et al. p16INK4a and frailty in hematologic malignancy. J Cancer Surviv . 2024. Doi:10.1007/s11764-024-01554-x Rosko AE, Huang Y, Wall SA, et al. CARG toxicity calculator in hematologic malignancy. J Geriatr Oncol . 2024. Doi:10.1016/j.jgo.2024.101702 Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging. Aging (Albany NY) . 2018;10(4):573-91. Doi:10.18632/aging.101414 Brownlee J. Data preparation for machine learning . Machine Learning Mastery; 2020. Hasan MK, Alam MA, Roy S, et al. Missing value imputation in machine learning. Inform Med Unlocked . 2021;27:100799. Doi:10.1016/j.imu.2021.100799 Vostokov D. Challenges of Python debugging in AI. In: Python debugging for AI, machine learning, and cloud computing . 2023. P. 199-212. Doi:10.1007/978-1-4842-9252-1_9 Zhao T, Zheng Y, Wu Z. Feature selection for nonlinear processes. Comput Chem Eng . 2023;169:108074. Doi:10.1016/j.compchemeng.2022.108074 Lindberg DS, Prosperi M, Bjarnadottir RI, et al. Machine learning for inpatient fall risk prediction. Int J Med Inform . 2020;143:104272. Doi:10.1016/j.ijmedinf.2020.104272 Liu Z, Kuo P-L, Horvath S, et al. Correction: A new aging measure. PloS Med . 2019;16(2):e1002760. Doi:10.1371/journal.pmed.1002760 Johnson AE, Zhu J, Garrard W, et al. Area deprivation index and cardiac readmissions. J Am Heart Assoc . 2021;10(13):e020466. Doi:10.1161/JAHA.120.020466 University of Wisconsin School of Medicine and Public Health. 2015 area deprivation index v2.0. [Internet]. Available from: https://www.neighborhoodatlas.medicine.wisc.edu/. Cited March 17, 2025. Knighton AJ, Savitz L, Belnap T, et al. Area deprivation index in population health. EGEMS . 2016;4(3):1238. Doi:10.13063/2327-9214.1238 Atkinson TM, Ryan SJ, Bennett AV, et al. CTCAE and patient-reported outcomes. Support Care Cancer . 2016;24:3669-76. Doi:10.1007/s00520-016-3297-9 Ahmad N, Rehman HSU, Malik MH, et al. Diagnostic prediction using machine learning. J Comput Biomed Inform . 2024. Doi:10.56907/jcbi.2024.0012 Osisanwo F, Akinsola J, Awodele O, et al. Supervised machine learning algorithms. Int J Comput Trends Technol . 2017;48(3):128-38. Mahajan P, Uddin S, Hajati F, et al. Machine learning ensemble approaches for disease prediction. Health Technol . 2024;14(3):597-613. Doi:10.1007/s12553-024-00823-0 Chiarito M, Stolfo D, Villaschi A, et al. Predicting survival in severe heart failure. Eur J Heart Fail . 2025. Doi:10.1002/ejhf.3045 Chan C, Lopez A, Castaneda G, et al. Black patients with colorectal cancer have advanced stage at diagnosis. J Community Health . 2017;42(4):724-9. Doi:10.1007/s10900-017-0312-0 Emeny RT, Carpenter DO, Lawrence DA. Health disparities: intracellular consequences. Toxicol Appl Pharmacol . 2021;416:115444. Doi:10.1016/j.taap.2021.115444 Hooten NN, Pacheco NL, Smith JT, et al. The accelerated aging phenotype. Ageing Res Rev . 2022;73:101536. Doi:10.1016/j.arr.2021.101536 Jessup JM, Stewart A, Greene FL, et al. Adjuvant chemotherapy for stage III colon cancer. JAMA . 2005;294(21):2703-11. Doi:10.1001/jama.294.21.2703 Smithson MG, McLeod MC, Al-Obaidi M, et al. Racial differences in aging-related deficits in colorectal cancer. Dis Colon Rectum . 2023;66(9):1245-53. Doi:10.1097/DCR.0000000000002533 Yost K, Perkins C, Cohen R, et al. Socioeconomic status and breast cancer incidence. Cancer Causes Control . 2001;12(8):703-11. Doi:10.1023/a:1011240019516 Eskander MF, Schapira EF, Bliss LA, et al. Marital status and cancer treatment outcomes. Am J Surg . 2016;212(4):691-9. Doi:10.1016/j.amjsurg.2016.06.010 Guven DC, Sahin TK, Erul E, et al. Pan-immune-inflammation value and cancer prognosis. Cancers (Basel) . 2022;14(11):2675. Doi:10.3390/cancers14112675 George M, Smith A, Sabesan S, et al. Physical comorbidities in older cancer patients. JMIR Cancer . 2021;7(4):e26425. Doi:10.2196/26425 Lashinger L, Rossi E, Hursting S. Obesity and resistance to chemotherapy. Clin Pharmacol Ther . 2014;96(4):458-63. Doi:10.1038/clpt.2014.136 Simkens LH, Koopman M, Mol L, et al. Body mass index and outcome in advanced colorectal cancer. Eur J Cancer . 2011;47(17):2560-7. Doi:10.1016/j.ejca.2011.07.026 Griggs JJ, Mangu PB, Anderson H, et al. Chemotherapy dosing for obese patients. J Clin Oncol . 2012;30(13):1553-61. Doi:10.1200/JCO.2011.39.9436 Drami I, Pring E, Gould L, et al. Body composition and chemotherapy dosing in colorectal cancer. Clin Oncol . 2021;33(12):e540-e52. Doi:10.1016/j.clon.2021.08.008 Vassbakk-Brovold K, Berntsen S, Fegran L, et al. Lifestyle intervention in patients undergoing chemotherapy. PloS One . 2015;10(7):e0131355. Doi:10.1371/journal.pone.0131355 Rambach L, Bertaut A, Vincent J, et al. Prognostic value of hematological toxicity in colorectal cancer. World J Gastroenterol . 2014;20(6):1565. Doi:10.3748/wjg.v20.i6.1565 Alexandre J, Gross-Goupil M, Falissard B, et al. Nutritional and inflammatory status in cancer patients. Ann Oncol . 2003;14(1):36-41. Doi:10.1093/annonc/mdg019 Baar MP, Brandt RM, Putavet DA, et al. Targeted apoptosis of senescent cells. Cell . 2017;169(1):132-47. Doi:10.1016/j.cell.2017.02.031 Merz LE, Story CM, Osei MA, et al. Absolute neutrophil count by Duffy status. Blood Adv . 2023;7(3):317-20. Doi:10.1182/bloodadvances.2022007754 Tables Tables 1 to 3 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files SuppTables13AIMLchemotoxicitymodelsCRC.docx Table123.docx Cite Share Download PDF Status: Published Journal Publication published 06 Oct, 2025 Read the published version in BMC Cancer → Version 1 posted Editorial decision: Revision requested 01 Jul, 2025 Reviews received at journal 18 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers agreed at journal 14 Jun, 2025 Reviews received at journal 04 Jun, 2025 Reviewers agreed at journal 23 May, 2025 Reviewers agreed at journal 23 May, 2025 Reviewers agreed at journal 23 May, 2025 Reviewers invited by journal 23 May, 2025 Editor assigned by journal 22 May, 2025 Editor invited by journal 13 May, 2025 Submission checks completed at journal 13 May, 2025 First submitted to journal 13 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-6628340\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":457279305,\"identity\":\"c0b3bbf4-5ed6-4827-9648-f20004da11ea\",\"order_by\":0,\"name\":\"Claire Han\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxElEQVRIiWNgGAWjYBACAxCRUGFTzw/hMxOp5cGZtATJBlK0MD5sO5xgcIBYLebsZw8wJLal5Rkfb38mwVBhndhASItlT14CQ8I5m2KzM2fMJBjOpBPWYnAgx4AhoSyNcduNHDYJxrbDRGg5/waohe0w4+YZ6c8kGP8Ro+UGyBag4RskEswkGBuI0vLG4EDCmTRjiTNnjC0SjqUbE+GwHMOHPyps5Pjb2x/e+FBjLUtQCwgcgLMSiFE+CkbBKBgFo4AwAABZxkJtzZCl8gAAAABJRU5ErkJggg==\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Claire\",\"middleName\":\"\",\"lastName\":\"Han\",\"suffix\":\"\"},{\"id\":457279306,\"identity\":\"ec7ee4cc-d43c-40df-bb01-708765af2782\",\"order_by\":1,\"name\":\"Christin Burd\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Christin\",\"middleName\":\"\",\"lastName\":\"Burd\",\"suffix\":\"\"},{\"id\":457279307,\"identity\":\"2a8068a3-44bf-4ebb-8d31-749c83572297\",\"order_by\":2,\"name\":\"Jesse Plascak\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Jesse\",\"middleName\":\"\",\"lastName\":\"Plascak\",\"suffix\":\"\"},{\"id\":457279308,\"identity\":\"c45be004-d780-4e6e-b5ac-db3228845fa8\",\"order_by\":3,\"name\":\"Fode Tounkara\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Fode\",\"middleName\":\"\",\"lastName\":\"Tounkara\",\"suffix\":\"\"},{\"id\":457279309,\"identity\":\"b95ee0ce-86ca-4b67-9d8d-8bcb47a0a5a1\",\"order_by\":4,\"name\":\"Ashley Rosko\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Ashley\",\"middleName\":\"\",\"lastName\":\"Rosko\",\"suffix\":\"\"},{\"id\":457279310,\"identity\":\"3486226d-9308-4925-94b8-480c7abe12e6\",\"order_by\":5,\"name\":\"Anne Noonan\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Anne\",\"middleName\":\"\",\"lastName\":\"Noonan\",\"suffix\":\"\"},{\"id\":457279311,\"identity\":\"7bfb805f-33fc-4ffe-af13-e5201941fb73\",\"order_by\":6,\"name\":\"Alai Tan\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Alai\",\"middleName\":\"\",\"lastName\":\"Tan\",\"suffix\":\"\"},{\"id\":457279312,\"identity\":\"92c51129-5add-433b-ad84-20f3fc72a398\",\"order_by\":7,\"name\":\"Diane Von Ah\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Diane\",\"middleName\":\"\",\"lastName\":\"Von Ah\",\"suffix\":\"\"},{\"id\":457279313,\"identity\":\"ec3a93ca-1996-4726-95c6-9704eb16c48d\",\"order_by\":8,\"name\":\"Xia Ning\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Ohio State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Xia\",\"middleName\":\"\",\"lastName\":\"Ning\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-05-09 11:53:24\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-6628340/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-6628340/v1\",\"draftVersion\":[],\"editorialEvents\":[{\"content\":\"https://doi.org/10.1186/s12885-025-14831-4\",\"type\":\"published\",\"date\":\"2025-10-06T15:57:01+00:00\"}],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":83012893,\"identity\":\"b30ad474-9bcf-46ae-9a63-0c7f044f3beb\",\"added_by\":\"auto\",\"created_at\":\"2025-05-19 05:41:25\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":390454,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eReceiver Operating Characteristic (ROC) Curves and Area Under the Curve (AUC) for Model Performance. The ROC curve is a graphical representation of a model’s diagnostic ability, plotting the True Positive Rate (TPR) against the False Positive Rate (FPR). Both graphs plot the True Positive Rate (TPR) on the y-axis, ranging from 0 to 1, and the False Positive Rate (FPR) on the x-axis, also ranging from 0 to 1. The AUC in ROC curves represents the probability that the classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one. An AUC close to 1 indicates a high model accuracy, 0.7 \\u003cu\\u003e\\u0026lt;\\u003c/u\\u003e AUC \\u0026lt; 0.8 indicates a good and moderate model accuracy, 0.5 \\u003cu\\u003e\\u0026lt;\\u003c/u\\u003e AUC \\u0026lt; 0.7 indicates a poor performance, and an AUC \\u0026lt; 0.5 indicates a fail. DT: Decision tree; GBM: Gradient boosting machine; LR: Logistic regression; RF: Random forest; SVM: Support vector machine; Gboosting = Gradient Boosting Machine; XGBoost: Extreme gradient boosting\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-6628340/v1/38baf35c8433d07606840f0c.png\"},{\"id\":83012892,\"identity\":\"ce07d4dd-531b-471b-90fe-b4213834d2cb\",\"added_by\":\"auto\",\"created_at\":\"2025-05-19 05:41:24\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":38578,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eFeature Importance\\u003c/strong\\u003e. The bar graph indicates the positive associations of input features with a better GI health condition. The height of the bar graph of the feature importance represents the importance of the feature, with relative importance being compared with the importance values of other features to understand their significance. The feature importance scores were based on the permutation feature importance (PFI) method. Interpretation of this graph should be based on a relative comparison of the values.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-6628340/v1/f158aeaa35d06be89f6e1dcc.png\"},{\"id\":93420784,\"identity\":\"167afbd0-8d3b-4855-9007-1ca6e4ea061a\",\"added_by\":\"auto\",\"created_at\":\"2025-10-13 16:10:27\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1278563,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-6628340/v1/4f003e52-badc-42ca-b97b-9f905e48a40e.pdf\"},{\"id\":83012311,\"identity\":\"eb4ab233-cb32-4dbe-b55e-fee95da89eee\",\"added_by\":\"auto\",\"created_at\":\"2025-05-19 05:33:25\",\"extension\":\"docx\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":57738,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"SuppTables13AIMLchemotoxicitymodelsCRC.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-6628340/v1/1976dbc95bd4b05dd7d88f63.docx\"},{\"id\":83012309,\"identity\":\"6ebc300f-1959-4534-944d-501c0a488b0a\",\"added_by\":\"auto\",\"created_at\":\"2025-05-19 05:33:24\",\"extension\":\"docx\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":35536,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Table123.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-6628340/v1/61950cca6f00ff772c6ff9c7.docx\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Machine Learning-based Chemotoxicity Predictions in Patients with Colorectal Cancer: Integrating Race, Geospatial Social Determinants of Health, and Biological Aging\",\"fulltext\":[{\"header\":\"INTRODUCTION\",\"content\":\"\\u003cp\\u003eColorectal cancer (CRC) remains a significant public health burden in the United States, with an estimated 153,020 new cases and 52,550 deaths projected for 2024 (1). Despite advances in screening and treatment, CRC is the third most commonly diagnosed cancer and the second leading cause of cancer-related mortality in the U.S. (1). 5-Fluorouracil (5-FU)-based chemotherapy is a cornerstone of CRC treatment, particularly for patients with stage II-III CRC (2). While 5-FU-based regimens have improved survival outcomes, they are associated with significant challenges, including severe chemotoxicity that can compromise treatment efficacy and patient quality of life in CRC (2-6). These side effects underscore the need for a deeper understanding of the factors influencing chemotoxicity and its management. Chemotherapy-induced toxicity is a major concern in CRC treatment, with up to 30-40% of patients experiencing severe adverse effects such as hematological toxicity (e.g., neutropenia, leucopenia), gastrointestinal (GI) toxicity (e.g., colitis, diarrhea, abdominal pain, gastrointestinal bleeding), neurotoxicity, and cardiotoxicity (7-10). Of note, GI and hematological complications are the most frequent and severe in CRC (7-10). These complications not only cause physical distress but also contribute to psychological burdens, including anxiety, depression, and worsened pre-existing stress (6). The prevalence of chemotherapy-related complications is alarmingly high, requiring dose reductions, treatment delays, or discontinuation due to toxicity (8, 11, 12). Such interruptions in therapy are associated with poorer survival outcomes, increased hospital admissions, and frequent emergency department visits, further straining healthcare resources (13). Additionally, chemotoxicity significantly impairs patients\\u0026apos; quality of life, affecting their ability to perform daily activities and maintain social relationships (14, 15). These challenges highlight the critical need for effective strategies to predict, prevent, and manage chemotherapy-related toxicity in CRC patients.\\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eDespite the widespread use of 5-FU-based chemotherapy, significant knowledge gaps persist in understanding and managing chemotoxicity. Previous studies have been limited by small sample sizes, lack of racial and ethnic diversity, and insufficient attention to social determinants of health (SDOH) that may influence toxicity risk in CRC (7-10, 16). Most research in chemotoxicity prediction in cancers has focused on non-Hispanic White populations with other cancer types (primarily breast and lung cancers), leaving underrepresented groups, such as African American/Black, Hispanic or Latino, Asian, and Indigenous patients, inadequately studied (17-19). Emerging evidence suggests that racialized groups and SDOH play a critical role in cancer health outcomes, such as chemotoxicity (20, 21). Studies have shown that racial and ethnic minoritized groups often experience higher rates of chemotherapy-related complications, which may be attributed to disparities in access to care, socioeconomic status, or other social factors, and underlying comorbidities (20, 21). Additionally, chronic stress, which is more prevalent in marginalized populations, has been linked to dysregulated immune function and increased susceptibility to chemotoxicity(22-24). For example, Black and Hispanic CRC patients are more likely to report severe symptom burden and impaired quality of life during chemotherapy compared to White patients (16). These disparities highlight the importance of incorporating race and SDOH into toxicity risk prediction models. Residential socioeconomic conditions should also be included and can be measured through validated measures relying on socioeconomic surveys of residents administered by the US Census Bureau (25, 26). Therefore, our study suggested integrating geospatial SDOH measured by the Area Deprivation Index (ADI) to understand the CRC population\\u0026apos;s chemotoxicity risk.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eFurthermore, emerging evidence shows that baseline biological aging is associated with immune functions, which predict chemotoxicity in cancer patients (27-29). Levine PhenoAge is a validated biological aging marker that can be computed using routine circulatory blood samples, including immune and inflammation markers, complete blood cell counts, and liver and kidney functions, without further blood assays (30). Therefore, baseline Levine PhenoAge may have the potential as a biological aging marker in predicting chemotoxicity, requiring further validation in CRC.\\u003c/p\\u003e\\n\\u003cp\\u003eGiven the limited understanding of risk factors of chemotoxicity mentioned above, there is a lack of reliable risk prediction tools, including comprehensive data to identify patients at high risk for chemotoxicity in CRC, which hinders proactive monitoring and personalized management strategies (8, 11). The current \\u0026quot;one-size-fits-all\\u0026quot; approach to chemotherapy management has proven inadequate, as it fails to account for the heterogeneity in toxicity profiles among patients. For instance, while some patients experience severe GI toxicity, others may develop hematological toxicity or other toxicities (e.g., neurotoxicity or cardiotoxicity), each requiring distinct management approaches (8, 11). This underscores the need for a more in-depth understanding of the types and mechanisms of chemotoxicity across racially and socially diverse and larger samples, to develop targeted interventions.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eTo address these gaps, our study leveraged the electronic health records (EHRs) including SDOH (e.g., geographic variations), from a large and racially diverse cohort of CRC patients. In our study, we applied artificial intelligence (AI)/machine learning (ML) methods that offer advantages over traditional statistical methods, such as multivariate regression, to develop predictive models for chemotoxicity in CRC patients. Unlike regression models, which assume linear relationships and struggle with high-dimensional data, AI/ML can capture complex, non-linear interactions among clinical, biological, and social factors (31-34). AI/ML also handles large, diverse datasets, such as EHRs, which include structured data (e.g., lab results) and unstructured data (e.g., clinical notes). Moreover, AI/ML outperforms traditional methods in predictive performance compared to regression-based approaches (31-34).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eTherefore, the aim of this study was to 1) develop AI/ML-based chemotoxicity prediction models (global, GI, and hematological chemotoxicity incidences) over the 6 months of 5-FU-based chemotherapy, and 2) identify the importance of chemotoxicity risk factors at baseline, including race, biological aging markers (i.e, Levine PhenoAge), and SDOH (e.g., geospatial variations measured by ADI)\\u003cstrong\\u003e.\\u0026nbsp;\\u003c/strong\\u003eWe hypothesized that our AI/ML models would demonstrate high accuracy of chemotoxicity prediction, baseline biological aging markers (Levine PhenoAge), ADI, and race would be significant predictors of chemotoxicity risk, and key risk factors may differ by types of chemotoxicity. By leveraging AI/ML, our study aims to develop more accurate and inclusive prediction models for chemotoxicity. This will enable proactive monitoring and personalized management, ultimately improving treatment outcomes and quality of life, and reducing disparities in CRC.\\u0026nbsp;\\u003c/p\\u003e\"},{\"header\":\"METHODS\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eStudy Design, Setting, and Data Sources\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis retrospective cohort study utilized EHR data from the Ohio State University (OSU) Comprehensive Cancer Center, covering the data from January 1, 2010, to December 31, 2020. The Ohio State University Comprehensive Cancer Center is a large not-for-profit academic cancer center in the Columbus, Ohio region with a catchment area encompassing the entire state. The OSU Honest Broker Operations Committee (HBOC) service has identified the eligible cohort and provided de-identified EHR data for our study in compliance with the US Health Insurance Portability and Accountability Act (HIPAA) (35). The study population included 1,735 adult patients diagnosed with stage II-III CRC who received 5-FU-based chemotherapy infusions. Inclusion Criteria include: Adult patients aged \\u003cu\\u003e\\u0026gt;\\u003c/u\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e18\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003ewith stage II-III CRC as the primary cancer site; underwent colon or rectal surgery without a stoma; scheduled for 8-12 cycles of 5-FU-based chemotherapy infusions (5-FU alone, Xeloda, or FOLFOX); single primary cancer; and having necessary baseline data including sociodemographic and clinical data, routine blood tests at baseline, and 6 months post-chemotherapy initiation, and zip codes were included. Exclusion Criteria include\\u003cu\\u003e:\\u003c/u\\u003e Patients having a current ostomy, chronic bowel disorders; regular use of steroids/immune suppressants; history of neoadjuvant chemotherapy; on active radiation or immune therapies as primary cancer therapies; and pregnant women; had incomplete or missing EHR data required for this study at baseline before chemotherapy and 6 months after chemotherapy were excluded.\\u003cstrong\\u003e\\u003c/strong\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eMeasures\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eD\\u003c/strong\\u003e\\u003cstrong\\u003eemographic and Clinical Dat\\u003c/strong\\u003e\\u003cstrong\\u003ea\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eDemographic data include race, ethnicity, chronological age, sex, and marital status. Clinical data include chemotherapy regimens, doses, and durations/cycles, weight, diet and exercise habits, smoking status, alcohol consumption, cancer stages, time since CRC diagnosis, history of radiotherapy, immunotherapy or surgery, Charlson Comorbidity Index, and blood-based immune and nutritional markers (e.g., albumin, platelets, hemoglobin). Among these variables, we controlled sex, cancer sites, and insurance types as covariates.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eBiological Aging Markers (Levine PhenoAge)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe used Levine PhenoAge to measure biological aging. In our study, the Levine PhenoAge was determined from routine monthly blood tests and included nine biomarkers in accordance with the method described previously. This method, validated in multiple studies including cancer patients, is based on the following measures (30, 36): Albumin (g/L) and Alkaline Phosphatase (U/L) \\u0026mdash;Liver function, Creatinine (umol/L)\\u0026mdash;Kidney function, Glucose (mmol/L)\\u0026mdash;Metabolism, C-reactive protein (CRP) (mg/dl)\\u0026mdash;Inflammation, Lymphocyte (%), Mean Cell Volume (MCV) (Fl), Red Cell Distribution Width (RCDW) (%), White Blood Cell Counts (WBC) (1K/ul)\\u0026mdash;Immune system, and Chronological age (years).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cimg src=\\\"data:image/png;base64,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\\\" width=\\\"811\\\" height=\\\"162\\\"\\u003e\\u003cbr\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cbr\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eSDOH, including geospatial variations (ADI)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe included SDOH-related variables from the EHR: employment status and geospatial variations measured by ADI (i.e., residential zip code-based socioeconomic and environmental deprivation). Other SDOH variables\\u0026mdash;like education levels, home ownership, annual household income, and poverty levels\\u0026mdash;were excluded due to a significant amount of missing data in the EHR. \\u003cem\\u003e\\u003cu\\u003eADI:\\u0026nbsp;\\u003c/u\\u003e\\u003c/em\\u003eThe ADI measures Residential zip code-based socioeconomic and environmental deprivation (ADI)\\u003cstrong\\u003e.\\u003c/strong\\u003e The ADI is a composite measure scored from 1 (least disadvantaged) to 100 (most disadvantaged), calculated using weighted coefficients from 17 indicators, such as income, education, employment, population age, poverty, and housing conditions (37). Developed by the University of Wisconsin, the ADI dataset is structured by ZIP codes +4 geographic areas (38). For this study, the 2015 ADI version was used to match patient ZIP codes +4 geographic areas available in the EHR, and ADI scores were divided into tertiles, with the highest tertile indicating the most socioeconomically disadvantaged group, following prior validation methods (37-39).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eOutcome\\u003c/strong\\u003e\\u003cstrong\\u003es\\u003c/strong\\u003e\\u003cstrong\\u003e: Chemotoxicity\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003ei\\u003c/strong\\u003e\\u003cstrong\\u003encidence\\u003c/strong\\u003e\\u003cstrong\\u003es over 6 months of 5-FU-based chemotherapy\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe assessed global, GI, and hematological chemotoxicity incidences that occurred from baseline to 6 months post-chemotherapy initiation. Chemotoxicity incidences were measured as dichotomized outcomes (presence versus absence) over the 6 months of chemotherapy using the clinician-reported adverse events. The chemotoxicity was reported by clinicians, based on laboratory or diagnostic data using the Common Terminology Criteria for Adverse Events criteria (CTCAE v.5.0: Grade I \\u0026lsquo;mild\\u0026rsquo;, II \\u0026lsquo;moderate\\u0026rsquo;, III \\u0026lsquo;severe\\u0026rsquo;, IV \\u0026lsquo;life-threatening\\u0026rsquo;, and V \\u0026lsquo;death related to adverse events\\u0026rsquo;) (reliability=0.85, sensitivity=0.79) (40). The OSU HBOC service used the International Classification of Diseases (ICD) codes 9 or 10 to identify chemotoxicity. \\u003cstrong\\u003eData Analyses\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eOverall Statistical Methods\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eDescriptive statistics were used to summarize the sociodemographic and clinical characteristics of the study population. A p-value of \\u0026lt; 0.05 was applied to determine statistical significance. We used R software, the MLR (Machine Learning in R) R package (version 3.6.3, R Foundation for Statistical Computing, Vienna, Austria), and Python (version 3.10.2, Python Software Foundation, Wilmington, U.S.) when appropriate for statistical analyses. Then, we evaluated associations of potential predictors (sociodemographic and clinical variables, Levine PhenoAge, and SDOH including geospatial variations) with chemotoxicity using the Chi-square test, or analysis of variance (ANOVA) for chemotoxicity incidences (as categorical variables) in whole datasets. We included three types of chemotoxicity incidences (global, GI, and hematological) as outcome variables, and performed all analyses separately. Given the purpose of our study to identify baseline predictors of chemotoxicity, we did not adjust for post-baseline measures of time-varying variables (i.e., blood-based immune markers and Levine PhenoAge). This step identified significant factors for inclusion in the AI/ML model training. Potential predictors were included only if they were significant in initial analyses.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAI/ML\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eApproach\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAI/ML\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eModeling.\\u0026nbsp;\\u003c/strong\\u003eAmong the 1,735 adult CRC patients, all had Levine PhenoAge data, and ZIP codes, and most of the features (i.e., risk factors) used were available with minimal missing data (missing data rates \\u0026lt; 5%). Given the low rates of missing data, we conducted complete case analyses. Outliers were identified and removed using the interquartile range (IQR), and duplicate entries were eliminated to prevent model skewing. Then, we performed data transformation, including normalization (scaling numerical features to a range of 0\\u0026ndash;1) to ensure equal contribution of all features to the models, and standardization (adjusting to a mean of 0 and a standard deviation of 1) to reduce feature dimensions and improve the performance and stability of our ML models [49]. Ordinal encoding was applied to categorical features with a natural order.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAI/ML Model\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eClassification\\u003c/strong\\u003e\\u003cstrong\\u003e\\u0026nbsp;and Evaluations.\\u0026nbsp;\\u003c/strong\\u003eWe first constructed a training dataset by selecting 80% of the CRC patients (n = 1,388) and reserving the remaining 20% (n = 347) for the test dataset. We matched age groups (10-year intervals), race, sex, and incidences of chemotoxicity (global, GI, and hematological toxicities) between training and test datasets. Using Python\\u0026rsquo;s `train_test_split` function from the scikit-learn (sklearn) library, samples were assigned to either the training or testing set (41). We also compared the comparability between training and test datasets (Table 1). Covariates were controlled for AI/ML modeling. The training dataset was used for initial model development, and the test dataset was employed to evaluate the model\\u0026rsquo;s performance in predicting chemotoxicity incidences. We examined the data compatibility between training and test datasets using Chi-square tests or ANOVA (Table 1). We applied six supervised ML methods: logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), gradient boosting machine (GBM), and extreme gradient boosting (XGBoost) (42). Each method was evaluated using a specific learning algorithm to assess its effectiveness in chemotoxicity predictions. Logistic regression, a widely used binary classifier, served as the baseline model for comparison. To prevent overfitting, hyperparameter tuning was performed using a random search with 5-fold cross-validation. The 5-fold cross-validation results were evaluated using the mean value of each model performance metric with a 95% confidence interval (CI) (42, 43). Model performance was assessed using metrics such as accuracy, precision, recall (sensitivity), specificity, F1 score, and area under the receiver operating characteristic curve (AUC) (42, 43).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003ePermutation Feature Importance (PFI)\\u003c/strong\\u003e\\u003cstrong\\u003e.\\u0026nbsp;\\u003c/strong\\u003eTo identify and rank the most significant risk factors for predicting chemotoxicity, a feature importance analysis was performed (44). Using PFI, the contribution of each feature was quantified by assessing its impact on the performance of the best-performing ML models. This involved randomly shuffling each feature 10 times individually within the trained classification model and measuring the resulting decline in model accuracy. The analysis utilized hold-out test data to evaluate the influence of features on chemotoxicity outcomes.\\u0026nbsp;\\u003c/p\\u003e\"},{\"header\":\"RESULTS\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eInitial Descriptive Analyses\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eTable 1 outlines the demographic and clinical characteristics, biological aging markers (Levine PhenoAge), SDOH including geospatial variations (ADI), and chemotoxicity outcomes for the training (n = 1,388, 80% of the total sample of 1,735 adults with CRC) and test (n = 347, 20% of the total sample) datasets. No significant differences were observed between datasets. The mean age of participants was 65.3 \\u0026plusmn; 13.2 years in the training dataset and 64.2 \\u0026plusmn; 12.6 years in the test dataset (p = 0.352). Levine PhenoAge was similar between the training (70.3) and test (70.1) datasets (p = 0.423). Approximately half of the participants were women in both datasets. The majority of patients were non-Hispanic White (67.2% in the training dataset and 72.0% in the test dataset), followed by non-Hispanic Black in both datasets (p = 0.984). We did not have Hispanic populations in our datasets. Most participants were married, had colon cancer, and were in Stage II, and had a high body mass index (BMI) \\u0026lsquo;overweight or obese\\u0026rsquo;, higher WBC and CRP levels than normal ranges, and a history of GI surgery. The neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) (higher levels indicating a pro-inflammatory status), were also similar, with the majority of patients within the normal range across the datasets. Most were married, insured (mainly private), employed, and had Stage II colon cancer. Chemotoxicity incidences were similar across datasets: global (56.3%), GI (40.7%), and hematological (22.8%) in total samples.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003ePotential Risk Factors for Chemotoxicity \\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eTable 2 summarizes and highlights potential risk factors significantly associated with chemotoxicity incidences, categorized into global, GI, and hematological chemotoxicity using an ANOVA or a Chi-square test. Chronological age was not associated with any type of chemotoxicity. Biological age was higher than chronological age in our datasets. Higher biological age (Levine PhenoAge) was a consistent risk factor across all chemotoxicity types. Non-Hispanic Black individuals were at higher risk for global and GI chemotoxicity, while non-Hispanic White individuals were at lower risk. Married/partnered marital status was negatively associated with chemotoxicity risk (Global and GI). Greater comorbidities were associated with global and GI chemotoxicity risk. Both underweight (BMI mean 17, range 15-22) and obesity (BMI \\u003cu\\u003e\\u0026gt;\\u003c/u\\u003e30) were risk factors for GI and global chemotoxicity, respectively. Elevated WBC, CRP, and NLR were strongly associated with increased risk for global and GI chemotoxicities, while lower WBC and NLR were associated with hematological toxicity risk. Previous cancer treatment histories, including radiation, immunotherapy, and GI surgery, were associated with higher global and GI chemotoxicities. In terms of health behaviors, heavy alcohol consumption increased GI chemotoxicity risk, while regular physical activity reduced global chemotoxicity risk. Higher ADI was linked to increased global and GI chemotoxicities, and unemployed or retired status influenced GI chemotoxicity risk. Detailed results are described in Supplementary Tables 1-3.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003ch4\\u003e\\u003cstrong\\u003ePerformance Comparison for Classification Models\\u003c/strong\\u003e\\u003c/h4\\u003e\\n\\u003cp\\u003eWe evaluated the performance of six ML models using a training dataset for development and a test dataset for validation, with five-fold cross-validation for global chemotoxicity, GI chemotoxicity, and hematological chemotoxicity (Table 3). For AI/ML chemotoxicity prediction modeling, we included potential risk factors showing significant associations with chemotoxicity in prior analyses (Table 2). For global chemotoxicity in the training dataset, SVM showed the highest AUC (0.988), accuracy (0.962), precision (0.954), sensitivity (0.971), specificity (0.953), and F1 score (0.963). In the test dataset, SVM also demonstrated the best performance with the highest AUC (0.987), accuracy (0.960), precision (0.949), sensitivity (0.971), specificity (0.948), and F1 score (0.960), indicating the superior ability to distinguish between positive and negative cases. For GI chemotoxicity in both training and test datasets, SVM achieved the highest AUC (0.984) in both datasets, as well as the highest values in other evaluation metrics, followed by XGBoost model, with overall good performance metrics shown in Table 3. For hematological chemotoxicity, similar results are shown; the SVM model achieved exceptional performance across all metrics in both training and test datasets, followed by the XGBoost model. The AI/ML prediction model AUC-ROC (Figure 1) also confirmed our findings in Table 3. \\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eFeature Importance\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe most significant features (based on PFI analyses) of the best-performing model (SVM in our study) in the training and test datasets were analyzed for each type of chemotoxicity. Among the significant input features identified in Table 2, the most important features for each type of chemotoxicity were overall consistent between the training and test datasets, except for employment status and previous history of cancer treatments. Figure 2 illustrates the relative importance of the input features of the AI/ML models.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e\\u003cu\\u003eGlobal Chemotoxicity\\u003c/u\\u003e\\u003c/em\\u003e\\u003cstrong\\u003e\\u003cem\\u003e.\\u0026nbsp;\\u003c/em\\u003e\\u003c/strong\\u003eThe SVM models in the training and the test dataset were evaluated separately using permutation feature importance (Figure 2A: orange bars represent the training dataset, and blue bars represent the test dataset). Significant features predicting global chemotoxicity in both datasets included being divorced/widowed or single, lack of physical activity, higher CRP levels, increased biological age (Levine PhenoAge), and higher ADI. Notably, being divorced/widowed or single and lack of physical activity were the most important features in the training dataset, while high CRP and being divorced/widowed or single were the most significant in the test dataset. Levine PhenoAge and ADI were consistently important features for global chemotoxicity in both datasets. Other contributing factors with minor impacts included greater comorbidities, higher NLR, racial minority status (primarily non-Hispanic Black), and obesity. In the training dataset, a previous history of cancer treatments was identified as a significant feature, while in the test dataset, being unemployed or retired was also significant, though both features had only a minor impact on global chemotoxicity.\\u003cem\\u003e\\u0026nbsp;\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e\\u003cu\\u003eGI Chemotoxicity\\u003c/u\\u003e\\u003c/em\\u003e\\u003cstrong\\u003e\\u003cem\\u003e.\\u0026nbsp;\\u003c/em\\u003e\\u003c/strong\\u003eThe SVM models in both training and test datasets were evaluated separately (Figure 2B). Significant features predicting GI chemotoxicity in both datasets included unemployed or retired status, high CRP levels, being divorced or single, higher biological age (Levine PhenoAge), higher ADI, higher WBC count, heavy alcohol consumption, higher NLR, racial/ethnic minority status (primarily non-Hispanic Black), greater comorbidities, lack of physical activity, and lower-than-normal BMI.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e\\u003cu\\u003eHematological Chemotoxicity\\u003c/u\\u003e\\u003c/em\\u003e\\u003cem\\u003e.\\u0026nbsp;\\u003c/em\\u003eThe SVM models in both training and test datasets were evaluated separately (Figure 2C). Significant features predicting hematological chemotoxicity in both datasets included lower WBC count, lower NLR, and higher biological age (Levine PhenoAge). Levine PhenoAge was a predictor of hematological toxicity, while ADI was not. Other SDOH variables did not contribute to predicting hematological toxicity.\\u003c/p\\u003e\"},{\"header\":\"DISCUSSION\",\"content\":\"\\u003cp\\u003eThis research is the first time to 1) develop AI/ML chemotoxicity prediction models including various common types of chemotoxicity (Global, GI, and hematological toxicities); 2) integrate a wide range of factors including biological aging and SDOH including geospatial variations; and 3) identify the most significant risk factors of chemotoxicity in patients with CRC. Additionally, our model used longitudinal EHR data, which improved chemotoxicity prediction. Our study demonstrated that the AI/ML algorithms successfully built prediction models with high accuracy, ranging from good (0.5 \\u0026le; AUC \\u0026lt; 0.7) to moderate to high (\\u0026gt;0.7)(33, 41, 42). The study also highlighted the relative importance of specific features in predicting chemotoxicity, revealing that biological age, race, and certain SDOH factors, including ADI, play a significant role in predicting chemotoxicity risk. The ML models developed and validated in this study have the potential to guide personalized strategies for identifying CRC patients at high risk of chemotoxicity, enabling tailored interventions to address unmet needs contributing to chemotoxicity risk in this population.\\u003c/p\\u003e\\n\\u003cp\\u003eIn our univariate analyses (ANOVA), chronological age was not associated with any types of chemotoxicities (Table 2). Conversely, biological age, as measured by Levine PhenoAge, was positively correlated with increased chemotoxicity across all categories (global, GI, and hematological) in univariate analyses as well as in AI/ML models. Biological age (measured using Levine PhenoAge) was higher than chronological age in our samples. Our study supports that biological aging, rather than chronological age alone, may be a more robust predictor of chemotoxicity risk. SDOH were significantly associated with chemotoxicity risk in our study. Biological aging, rather than chronological age, was a stronger predictor of chemotoxicity. SDOH, including higher ADI scores, unemployment or retirement, and being unmarried, were linked to greater global and GI toxicities, underscoring the impact of socioeconomic and social support factors. Non-Hispanic Black individuals experienced higher toxicity rates, likely reflecting broader structural racial inequities, including access to healthcare, socioeconomic position, and neighborhood disadvantage (45-50). However, our study sample lacked representation of other racial and ethnic minority groups, such as Hispanic populations, underscoring the need for future research in more diverse cohorts to better understand the complex relationships between SDOH and cancer treatment outcomes (51).\\u003c/p\\u003e\\n\\u003cp\\u003eOur findings added new knowledge to the existing literature (52, 53) by providing evidence that pro-inflammatory status (high WBC, CRP, and NLR) (52) and greater comorbidity (53) are associated with cancer health outcomes, including chemotoxicity (specifically global and GI toxicity as well). Our findings highlight the role of pre-existing health conditions in exacerbating chemotoxicity. These findings suggest that systemic inflammation may serve as a key mediator of chemotoxicity, potentially exacerbating tissue damage and impairing recovery (52). Interventions targeting inflammation, such as anti-inflammatory therapies or lifestyle modifications, may help mitigate these risks.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eIn our study, BMI exhibited a complex relationship with chemotoxicity. While higher BMI (\\u0026gt;30 kg/m\\u0026sup2;) was associated with increased global toxicity, lower BMI (\\u0026lt;25 kg/m\\u0026sup2;) was linked to higher GI toxicity, particularly in underweight individuals with a mean BMI of 17 (range: 15\\u0026ndash;22). This underscores the potential dual role of BMI as both a protective and a risk factor of chemotoxicity. The complex relationship between high BMI and chemotoxicity can be explained as follows. For higher BMI and increased global toxicity, obesity may alter drug pharmacokinetics, increase systemic inflammation, and exacerbate comorbidities, leading to higher global toxicity (54, 55). Obese patients are also often prescribed large doses of chemotherapy, leading to excessive toxic effects (56). Low BMI, especially in underweight individuals, reflects poor nutritional status, reduced physiological reserve, and cachexia, making the GI tract more vulnerable to chemotherapy damage (57). Personalized chemotherapy dosing and supportive care (e.g., nutritional support for underweight patients and inflammation management for obese patients) are needed to mitigate toxicity risks with BMI.\\u003c/p\\u003e\\n\\u003cp\\u003ePrevious history of cancer treatments, including radiation, immunotherapy, and GI tract surgery, were associated with increased global and GI toxicity. This suggests that cumulative treatment burden may heighten susceptibility to chemotoxicity, particularly in patients with a history of aggressive or multimodal therapies. These findings highlight the importance of considering treatment history when assessing chemotoxicity risk and designing personalized treatment plans. Lifestyle behaviors, such as heavy alcohol use, were associated with increased GI toxicity, while regular physical activity was linked to reduced global toxicity. These findings underscore the potential of lifestyle interventions, such as alcohol reduction and exercise programs, in mitigating chemotoxicity risk (58). Integrating behavioral support into cancer care may improve treatment tolerance and long-term outcomes (58).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp; \\u0026nbsp;\\u0026nbsp;We found that greater biological age and lower WBC and NLR levels were significant risk factors for hematological toxicity, while clinical and SDOH factors were not. Similar findings were observed in other studies (8, 59, 60). This suggests hematological toxicity is primarily driven by intrinsic aging and immune/inflammation-related biological factors, such as reduced bone marrow reserve and immune capacity, rather than external clinical or socioeconomic influences (8, 59-61). Notably, the association between low NLR and toxicity should be interpreted with caution, as genetic variants like the Duffy-null phenotype\\u0026mdash;common in individuals of African descent\\u0026mdash;can result in lower ANC despite normal immune function, potentially influencing NLR values (62).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eClinical Implications.\\u0026nbsp;\\u003c/strong\\u003eThe identification of these risk factors has important clinical implications. AI/ML models incorporating biosocial markers could help identify CRC patients at high risk for chemotoxicity, enabling early interventions and personalized treatment strategies. For instance, patients with elevated inflammatory markers or high ADI scores may benefit from closer monitoring and targeted supportive care. Additionally, addressing modifiable risk factors, such as lifestyle behaviors and comorbidities, including obesity or malnutritional status, may improve treatment tolerance and quality of life. Lastly, oncologists should screen high-risk groups for different types of chemotoxicity (e.g., GI vs. hematological) and provide tailored interventions specific to each chemotoxicity type.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eStrengths and Limitations.\\u0026nbsp;\\u003c/strong\\u003eOur study leverages diverse input data, including biological aging and immune markers, and is not limited to non-Hispanic White samples. It incorporates longitudinal data, objective SDOH measures (e.g., ADI), and a large EHR dataset, enhancing the rigor of AI/ML modeling. However, as a secondary analysis of EHR data, our study may lack key variables, such as representation of other racial/ethnic minorities (e.g., Hispanic populations) and additional SDOH factors (e.g., income, home ownership, poverty levels, occupation, healthcare access, and social support). Additionally, while we controlled for changes in blood markers including Levine PhenoAge over time in our analyses, we did not examine their predictive role in chemotoxicity. Future longitudinal studies with multiple time points are needed to better understand the longitudinal predictability of changes in biological aging over time for chemotoxicity.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eFuture research should focus on expanding racial/ethnic diversity, incorporating comprehensive SDOH measures, and exploring longitudinal dynamics of time-varying factors including immune and biological aging markers, to refine chemotoxicity prediction models. External validation of our study using different datasets is needed to confirm our AI/ML models. Additionally, further research to investigate underlying biosocial mechanisms of chemotoxicity and tailored intervention strategies, is suggested.\\u0026nbsp;\\u003c/p\\u003e\"},{\"header\":\"CONCLUSION\",\"content\":\"\\u003cp\\u003eThe AI/ML chemotoxicity prediction models demonstrated high accuracy using SVM and XGBoost algorithms. This study highlights key predictors of chemotoxicity, including marital status (divorced/widowed or single), higher biological age (Levine PhenoAge), increased immune and inflammatory markers (e.g., WBC, CRP, NLR), and SDOH disparities such as geospatial variations (higher ADI) and employment status (unemployed or retired). Non-Hispanic Black race and BMI also play roles, with higher BMI associated with global toxicities and lower BMI with GI toxicities. Lower WBC, NLR, and higher biological age (Levine PhenoAge) predict greater hematological chemotoxicity. Risk factors vary by type of chemotoxicity (global, GI, and hematological). Findings emphasize the need for personalized treatment, addressing disparities, and targeted interventions (e.g., reducing inflammation, anti-aging strategies, and improving health disparities). These findings underscore the importance of a holistic approach to cancer care that addresses biological and social health determinants. By integrating these factors into predictive models, clinicians can better identify at-risk CRC patients and implement tailored interventions to mitigate chemotoxicity and improve survivorship outcomes.\\u003c/p\\u003e\"},{\"header\":\"Abbreviations\",\"content\":\"\\u003ctable border=\\\"0\\\" cellspacing=\\\"0\\\" cellpadding=\\\"0\\\" align=\\\"\\\"\\u003e\\n \\u003ctbody\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eAbbreviation\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eFull Term\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003e5-FU\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003e5-Fluorouracil\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eADI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eArea Deprivation Index\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eAI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eArtificial Intelligence\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eAUC\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eArea Under the Curve\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eBMI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eBody Mass Index\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eCI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eConfidence Interval\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eCRC\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eColorectal Cancer\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 201px;\\\"\\u003e\\n \\u003cp\\u003eCRP\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 338px;\\\"\\u003e\\n \\u003cp\\u003eC-Reactive Protein\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eCTCAE\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eCommon Terminology Criteria for Adverse Events\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eDT\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eDecision Tree\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eEHR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eElectronic Health Records\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eF1\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eF1 Score (harmonic mean of precision and recall)\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eGBM\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eGradient Boosting Machine\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eGI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eGastrointestinal\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eICD\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eInternational Classification of Diseases\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eIQR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eInterquartile Range\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eLR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eLogistic Regression\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eML\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eMachine Learning\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eNLR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eNeutrophil-to-Lymphocyte Ratio\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eOSU\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eOhio State University\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003ePFI\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003ePermutation Feature Importance\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003ePLR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003ePlatelet-to-Lymphocyte Ratio\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eR\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eStatistical software R\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eRCDW\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eRed Cell Distribution Width\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eRF\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eRandom Forest\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eSDOH\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eSocial Determinants of Health\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eSVM\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eSupport Vector Machine\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eWBC\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eWhite Blood Cell count\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eXGBoost\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\"\\u003e\\n \\u003cp\\u003eExtreme Gradient Boosting\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/tbody\\u003e\\n\\u003c/table\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003eThis study is a secondary data analysis using de-identified EHR data, Thus, the IRB approval was waived for the current study. IRB Ohio State University\\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eSupplementary Materials:\\u0026nbsp;\\u003c/strong\\u003eThe following supporting information (Supplementary Tables 1-3) can be downloaded at the journal site.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAuthor Contributions:\\u003c/strong\\u003e Conceptualization, CH, XN, CB, JP, AN, DV.; methodology, CH, XN, TF, AT; validation, CH, XN, CB, JP, FT, AT, DV; formal analysis, CH, FT.; investigation, CH, XB, CB, JP, AR, AT, DV; data curation, CH, FT.; writing\\u0026mdash;original draft preparation, CH; writing\\u0026mdash;review and editing, CH, XN, CB, JP, AN, AR, AT, DV; visualization, CH, FT.; supervision, XN. All authors have read and agreed to the published version of the manuscript.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eFunding:\\u003c/strong\\u003e The work was original research that had not been published previously.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003eCH\\u003c/em\\u003e: Cancer Research Seed Grant from the Ohio State University College of Nursing and the Ohio State University Comprehensive Cancer Center. CH is also funded by the Oncology Nurse Foundation (ONF) RE03. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eInstitutional Review Board Statement:\\u0026nbsp;\\u003c/strong\\u003eThank you for your feedback. Our study utilized de-identified secondary electronic health record (EHR) data from the Ohio State University Comprehensive Cancer Research Center. As this research involved retrospective analysis of pre-existing, anonymized data, institutional review board (IRB) approval was not required. The Ohio State University Cancer Center\\u0026rsquo;s IRB classified this study as non-human subjects research and granted an exemption per institutional policy and U.S. regulations (45 CFR 46). A waiver of informed consent was also granted. All data were fully anonymized and handled in compliance with HIPAA\\u0026rsquo;s Privacy Rule. Access to the dataset was governed by Ohio State University Comprehensive Cancer Research Center\\u0026rsquo;s data governance policies and a formal data-use agreement.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eInformed Consent Statement:\\u0026nbsp;\\u003c/strong\\u003eNot applicable.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eData Availability Statement:\\u003c/strong\\u003e The datasets analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request. Our study did not generate any data and we used the de-identified EHR data as a secondary data analysis. Furthermore, the de-identified dataset used in this study contains restricted geographic (zip code) and demographic (age \\u0026gt;80) elements that may pose re-identification risks under HIPAA. To comply with institutional and regulatory requirements, access is limited to qualified researchers through a formal data request process overseen by OSU, subject to a Data Use Agreement and institutional review.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAcknowledgments:\\u003c/strong\\u003e Dr. Andrey Loginov, PhD, Bioinformatician, AI/ML Engineer, and Big Data Scientist, University of Maryland School of Medicine, Baltimore, Maryland, United States. We thank Dr. Loginov for his assistance in data curation and AI/ML modeling.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eConflicts of Interest:\\u003c/strong\\u003e The authors declare no conflicts of interest.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col start=\\\"1\\\" type=\\\"1\\\"\\u003e\\n \\u003cli\\u003eSiegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. \\u003cem\\u003eCA Cancer J Clin\\u003c/em\\u003e. 2024;74(1). Doi:10.3322/caac.21820\\u003c/li\\u003e\\n \\u003cli\\u003eChan GHJ, Chee CE. Making sense of adjuvant chemotherapy in colorectal cancer. \\u003cem\\u003eJ Gastrointest Oncol\\u003c/em\\u003e. 2019;10(6):1183-92. Doi:10.21037/jgo.2019.09.03\\u003c/li\\u003e\\n \\u003cli\\u003eBiller LH, Schrag D. Diagnosis and treatment of metastatic colorectal cancer: a review. \\u003cem\\u003eJAMA\\u003c/em\\u003e. 2021;325(7):669-85. Doi:10.1001/jama.2021.0106\\u003c/li\\u003e\\n \\u003cli\\u003eExtermann M. Chemotherapy toxicity. In: Gu D, Dupre ME, editors. \\u003cem\\u003eEncyclopedia of gerontology and population aging\\u003c/em\\u003e. Cham: Springer; 2020. P. 1-6. Doi:10.1007/978-3-319-69892-2_1001-1\\u003c/li\\u003e\\n \\u003cli\\u003eSargent DJ, Goldberg RM, Jacobson SD, et al. A pooled analysis of adjuvant chemotherapy for resected colon cancer in elderly patients. \\u003cem\\u003eN Engl J Med\\u003c/em\\u003e. 2001;345(15):1091-7. Doi:10.1056/NEJMoa010957\\u003c/li\\u003e\\n \\u003cli\\u003eTantoy IY, Cataldo JK, Aouizerat BE, et al. A review of the literature on multiple co-occurring symptoms in patients with colorectal cancer. \\u003cem\\u003eCancer Nurs\\u003c/em\\u003e. 2016;39(6):437-45. Doi:10.1097/NCC.0000000000000336\\u003c/li\\u003e\\n \\u003cli\\u003eHan C, Yang G, Syrjala K. Symptom experiences in colorectal cancer survivors: a systematic review and meta-analysis. \\u003cem\\u003eCancer Nurs\\u003c/em\\u003e. 2020; 43(3):E132-E158. Doi:10.1097/NCC.0000000000000678\\u003c/li\\u003e\\n \\u003cli\\u003eHan CJ, Ning X, Burd CE, et al. Chemotoxicity and associated risk factors in colorectal cancer: a systematic review. \\u003cem\\u003eCancers (Basel)\\u003c/em\\u003e. 2024;16(14). Doi:10.3390/cancers16142662\\u003c/li\\u003e\\n \\u003cli\\u003eHan CJ, Reding KW, Kalady MF, et al. Factors associated with long-term gastrointestinal symptoms in colorectal cancer survivors. \\u003cem\\u003ePloS One\\u003c/em\\u003e. 2023;18(5):e0286058. Doi:10.1371/journal.pone.0286058\\u003c/li\\u003e\\n \\u003cli\\u003eGustavsson B, Carlsson G, Machover D, et al. A review of systemic chemotherapy in colorectal cancer. \\u003cem\\u003eClin Colorectal Cancer\\u003c/em\\u003e. 2015;14(1):1-10. Doi:10.1016/j.clcc.2014.11.002\\u003c/li\\u003e\\n \\u003cli\\u003eFeliu J, Espinosa E, Basterretxea L, et al. Prediction of chemotoxicity in older patients with colorectal cancer. \\u003cem\\u003eCancers (Basel)\\u003c/em\\u003e. 2022;14(1):127. Doi:10.3390/cancers14010127\\u003c/li\\u003e\\n \\u003cli\\u003eYazbeck V, Alesi E, Myers J, et al. An overview of chemotoxicity and radiation toxicity in cancer therapy. \\u003cem\\u003eAdv Cancer Res\\u003c/em\\u003e. 2022;155:1-27. Doi:10.1016/bs.acr.2022.02.001\\u003c/li\\u003e\\n \\u003cli\\u003eLyman GH, Yau L, Nakov R, et al. Overall survival and risk of second malignancies with chemotherapy and G-CSF. \\u003cem\\u003eAnn Oncol\\u003c/em\\u003e. 2018;29(9):1903-10. Doi:10.1093/annonc/mdy244\\u003c/li\\u003e\\n \\u003cli\\u003eAndreyev J, Adams R, Bornschein J, et al. British Society of Gastroenterology guidance on gastrointestinal complications of cancer treatment. \\u003cem\\u003eGut\\u003c/em\\u003e. 2025. [Epub ahead of print]. Doi:10.1136/gutjnl-2024-332115\\u003c/li\\u003e\\n \\u003cli\\u003eBaltussen JC, de Glas NA, van Holstein Y, et al. Chemotherapy-related toxic effects and quality of life in older patients. \\u003cem\\u003eJAMA Netw Open\\u003c/em\\u003e. 2023;6(10):e2339116. Doi:10.1001/jamanetworkopen.2023.39116\\u003c/li\\u003e\\n \\u003cli\\u003eHan CJ, Tounkara F, Kalady MF, et al. Racial/ethnic disparities in HRQoL in colorectal cancer survivors. \\u003cem\\u003eJ Gastrointest Cancer\\u003c/em\\u003e. 2024;55(3):1179-89. Doi:10.1007/s12029-024-01024-8\\u003c/li\\u003e\\n \\u003cli\\u003eAlibhai SM, Breunis H, Gregg RW, et al. Validating the CARG toxicity tool in prostate cancer. \\u003cem\\u003eJ Clin Oncol\\u003c/em\\u003e. 2019;37(15_suppl):e23081. Doi:10.1200/JCO.2019.37.15_suppl.e23081\\u003c/li\\u003e\\n \\u003cli\\u003eMoth EB, Kiely BE, Stefanic N, et al. Predicting chemotherapy toxicity in older adults. \\u003cem\\u003eJ Geriatr Oncol\\u003c/em\\u003e. 2019;10(2):202-9. Doi:10.1016/j.jgo.2018.08.004\\u003c/li\\u003e\\n \\u003cli\\u003ePathak N, Nishijima TF, Cavdar E. The applicability of the CARG chemotherapy toxicity tool. \\u003cem\\u003eJ Geriatr Oncol\\u003c/em\\u003e. 2024;15(8):102070. Doi:10.1016/j.jgo.2023.102070\\u003c/li\\u003e\\n \\u003cli\\u003eNipp RD, Shui AM, Perez GK, et al. Health care access among cancer survivors. \\u003cem\\u003eJAMA Oncol\\u003c/em\\u003e. 2018;4(6):791-7. Doi:10.1001/jamaoncol.2018.0097\\u003c/li\\u003e\\n \\u003cli\\u003eLorentsen MK, Sanoff HK. Social determinants of health and colorectal cancer outcomes. \\u003cem\\u003eCurr Treat Options Oncol\\u003c/em\\u003e. 2024;25(4):453-64. Doi:10.1007/s11864-024-01194-4\\u003c/li\\u003e\\n \\u003cli\\u003eHarris AR, Pichardo CM, Franklin J, et al. Multilevel stressors and tumor immunity in breast cancer. \\u003cem\\u003eJAMA Netw Open\\u003c/em\\u003e. 2025;8(2):e2459754. Doi:10.1001/jamanetworkopen.2024.59754\\u003c/li\\u003e\\n \\u003cli\\u003eGoel N, Hernandez A, Cole SW. Social genomic determinants of health. \\u003cem\\u003eJ Clin Oncol\\u003c/em\\u003e. 2024;42(30):3618-27. Doi:10.1200/JCO.23.02145\\u003c/li\\u003e\\n \\u003cli\\u003eTran T, Rousseau MA, Farris DP, et al. Social vulnerability index in cancer research. \\u003cem\\u003eCancer Causes Control\\u003c/em\\u003e. 2023;34(5):407-20. Doi:10.1007/s10552-023-01677-z\\u003c/li\\u003e\\n \\u003cli\\u003ePearson J, Jacobson C, Ugochukwu N, et al. Geospatial analysis of social determinants of health. \\u003cem\\u003eInt Anesthesiol Clin\\u003c/em\\u003e. 2023;61(1):49-62. Doi:10.1097/AIA.0000000000000389\\u003c/li\\u003e\\n \\u003cli\\u003eUS Census Bureau. American Community Survey, 2009-2013. [Internet]. Available from:https://data.census.gov/. Accessed March 17, 2025.\\u003c/li\\u003e\\n \\u003cli\\u003eFell G, Rosko AE, Abel GA, et al. Peripheral blood CD3+ T-cell biomarkers in frailty. \\u003cem\\u003eBr J Haematol\\u003c/em\\u003e. 2022;199(1):100-5. Doi:10.1111/bjh.18340\\u003c/li\\u003e\\n \\u003cli\\u003eRosko AE, Elsaid MI, Woyach J, et al. p16INK4a and frailty in hematologic malignancy. \\u003cem\\u003eJ Cancer Surviv\\u003c/em\\u003e. 2024. Doi:10.1007/s11764-024-01554-x\\u003c/li\\u003e\\n \\u003cli\\u003eRosko AE, Huang Y, Wall SA, et al. CARG toxicity calculator in hematologic malignancy. \\u003cem\\u003eJ Geriatr Oncol\\u003c/em\\u003e. 2024. Doi:10.1016/j.jgo.2024.101702\\u003c/li\\u003e\\n \\u003cli\\u003eLevine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging. \\u003cem\\u003eAging (Albany NY)\\u003c/em\\u003e. 2018;10(4):573-91. Doi:10.18632/aging.101414\\u003c/li\\u003e\\n \\u003cli\\u003eBrownlee J. \\u003cem\\u003eData preparation for machine learning\\u003c/em\\u003e. Machine Learning Mastery; 2020.\\u003c/li\\u003e\\n \\u003cli\\u003eHasan MK, Alam MA, Roy S, et al. Missing value imputation in machine learning. \\u003cem\\u003eInform Med Unlocked\\u003c/em\\u003e. 2021;27:100799. Doi:10.1016/j.imu.2021.100799\\u003c/li\\u003e\\n \\u003cli\\u003eVostokov D. Challenges of Python debugging in AI. In: \\u003cem\\u003ePython debugging for AI, machine learning, and cloud computing\\u003c/em\\u003e. 2023. P. 199-212. Doi:10.1007/978-1-4842-9252-1_9\\u003c/li\\u003e\\n \\u003cli\\u003eZhao T, Zheng Y, Wu Z. Feature selection for nonlinear processes. \\u003cem\\u003eComput Chem Eng\\u003c/em\\u003e. 2023;169:108074. Doi:10.1016/j.compchemeng.2022.108074\\u003c/li\\u003e\\n \\u003cli\\u003eLindberg DS, Prosperi M, Bjarnadottir RI, et al. Machine learning for inpatient fall risk prediction. \\u003cem\\u003eInt J Med Inform\\u003c/em\\u003e. 2020;143:104272. Doi:10.1016/j.ijmedinf.2020.104272\\u003c/li\\u003e\\n \\u003cli\\u003eLiu Z, Kuo P-L, Horvath S, et al. Correction: A new aging measure. \\u003cem\\u003ePloS Med\\u003c/em\\u003e. 2019;16(2):e1002760. Doi:10.1371/journal.pmed.1002760\\u003c/li\\u003e\\n \\u003cli\\u003eJohnson AE, Zhu J, Garrard W, et al. Area deprivation index and cardiac readmissions. \\u003cem\\u003eJ Am Heart Assoc\\u003c/em\\u003e. 2021;10(13):e020466. Doi:10.1161/JAHA.120.020466\\u003c/li\\u003e\\n \\u003cli\\u003eUniversity of Wisconsin School of Medicine and Public Health. 2015 area deprivation index v2.0. [Internet]. Available from: https://www.neighborhoodatlas.medicine.wisc.edu/. Cited March 17, 2025.\\u003c/li\\u003e\\n \\u003cli\\u003eKnighton AJ, Savitz L, Belnap T, et al. Area deprivation index in population health. \\u003cem\\u003eEGEMS\\u003c/em\\u003e. 2016;4(3):1238. Doi:10.13063/2327-9214.1238\\u003c/li\\u003e\\n \\u003cli\\u003eAtkinson TM, Ryan SJ, Bennett AV, et al. CTCAE and patient-reported outcomes. \\u003cem\\u003eSupport Care Cancer\\u003c/em\\u003e. 2016;24:3669-76. Doi:10.1007/s00520-016-3297-9\\u003c/li\\u003e\\n \\u003cli\\u003eAhmad N, Rehman HSU, Malik MH, et al. Diagnostic prediction using machine learning. \\u003cem\\u003eJ Comput Biomed Inform\\u003c/em\\u003e. 2024. Doi:10.56907/jcbi.2024.0012\\u003c/li\\u003e\\n \\u003cli\\u003eOsisanwo F, Akinsola J, Awodele O, et al. Supervised machine learning algorithms. \\u003cem\\u003eInt J Comput Trends Technol\\u003c/em\\u003e. 2017;48(3):128-38.\\u003c/li\\u003e\\n \\u003cli\\u003eMahajan P, Uddin S, Hajati F, et al. Machine learning ensemble approaches for disease prediction. \\u003cem\\u003eHealth Technol\\u003c/em\\u003e. 2024;14(3):597-613. Doi:10.1007/s12553-024-00823-0\\u003c/li\\u003e\\n \\u003cli\\u003eChiarito M, Stolfo D, Villaschi A, et al. Predicting survival in severe heart failure. \\u003cem\\u003eEur J Heart Fail\\u003c/em\\u003e. 2025. Doi:10.1002/ejhf.3045\\u003c/li\\u003e\\n \\u003cli\\u003eChan C, Lopez A, Castaneda G, et al. Black patients with colorectal cancer have advanced stage at diagnosis. \\u003cem\\u003eJ Community Health\\u003c/em\\u003e. 2017;42(4):724-9. Doi:10.1007/s10900-017-0312-0\\u003c/li\\u003e\\n \\u003cli\\u003eEmeny RT, Carpenter DO, Lawrence DA. Health disparities: intracellular consequences. \\u003cem\\u003eToxicol Appl Pharmacol\\u003c/em\\u003e. 2021;416:115444. Doi:10.1016/j.taap.2021.115444\\u003c/li\\u003e\\n \\u003cli\\u003eHooten NN, Pacheco NL, Smith JT, et al. The accelerated aging phenotype. \\u003cem\\u003eAgeing Res Rev\\u003c/em\\u003e. 2022;73:101536. Doi:10.1016/j.arr.2021.101536\\u003c/li\\u003e\\n \\u003cli\\u003eJessup JM, Stewart A, Greene FL, et al. Adjuvant chemotherapy for stage III colon cancer. \\u003cem\\u003eJAMA\\u003c/em\\u003e. 2005;294(21):2703-11. Doi:10.1001/jama.294.21.2703\\u003c/li\\u003e\\n \\u003cli\\u003eSmithson MG, McLeod MC, Al-Obaidi M, et al. Racial differences in aging-related deficits in colorectal cancer. \\u003cem\\u003eDis Colon Rectum\\u003c/em\\u003e. 2023;66(9):1245-53. Doi:10.1097/DCR.0000000000002533\\u003c/li\\u003e\\n \\u003cli\\u003eYost K, Perkins C, Cohen R, et al. Socioeconomic status and breast cancer incidence. \\u003cem\\u003eCancer Causes Control\\u003c/em\\u003e. 2001;12(8):703-11. Doi:10.1023/a:1011240019516\\u003c/li\\u003e\\n \\u003cli\\u003eEskander MF, Schapira EF, Bliss LA, et al. Marital status and cancer treatment outcomes. \\u003cem\\u003eAm J Surg\\u003c/em\\u003e. 2016;212(4):691-9. Doi:10.1016/j.amjsurg.2016.06.010\\u003c/li\\u003e\\n \\u003cli\\u003eGuven DC, Sahin TK, Erul E, et al. Pan-immune-inflammation value and cancer prognosis. \\u003cem\\u003eCancers (Basel)\\u003c/em\\u003e. 2022;14(11):2675. Doi:10.3390/cancers14112675\\u003c/li\\u003e\\n \\u003cli\\u003eGeorge M, Smith A, Sabesan S, et al. Physical comorbidities in older cancer patients. \\u003cem\\u003eJMIR Cancer\\u003c/em\\u003e. 2021;7(4):e26425. Doi:10.2196/26425\\u003c/li\\u003e\\n \\u003cli\\u003eLashinger L, Rossi E, Hursting S. Obesity and resistance to chemotherapy. \\u003cem\\u003eClin Pharmacol Ther\\u003c/em\\u003e. 2014;96(4):458-63. Doi:10.1038/clpt.2014.136\\u003c/li\\u003e\\n \\u003cli\\u003eSimkens LH, Koopman M, Mol L, et al. Body mass index and outcome in advanced colorectal cancer. \\u003cem\\u003eEur J Cancer\\u003c/em\\u003e. 2011;47(17):2560-7. Doi:10.1016/j.ejca.2011.07.026\\u003c/li\\u003e\\n \\u003cli\\u003eGriggs JJ, Mangu PB, Anderson H, et al. Chemotherapy dosing for obese patients. \\u003cem\\u003eJ Clin Oncol\\u003c/em\\u003e. 2012;30(13):1553-61. Doi:10.1200/JCO.2011.39.9436\\u003c/li\\u003e\\n \\u003cli\\u003eDrami I, Pring E, Gould L, et al. Body composition and chemotherapy dosing in colorectal cancer. \\u003cem\\u003eClin Oncol\\u003c/em\\u003e. 2021;33(12):e540-e52. Doi:10.1016/j.clon.2021.08.008\\u003c/li\\u003e\\n \\u003cli\\u003eVassbakk-Brovold K, Berntsen S, Fegran L, et al. Lifestyle intervention in patients undergoing chemotherapy. \\u003cem\\u003ePloS One\\u003c/em\\u003e. 2015;10(7):e0131355. Doi:10.1371/journal.pone.0131355\\u003c/li\\u003e\\n \\u003cli\\u003eRambach L, Bertaut A, Vincent J, et al. Prognostic value of hematological toxicity in colorectal cancer. \\u003cem\\u003eWorld J Gastroenterol\\u003c/em\\u003e. 2014;20(6):1565. Doi:10.3748/wjg.v20.i6.1565\\u003c/li\\u003e\\n \\u003cli\\u003eAlexandre J, Gross-Goupil M, Falissard B, et al. Nutritional and inflammatory status in cancer patients. \\u003cem\\u003eAnn Oncol\\u003c/em\\u003e. 2003;14(1):36-41. Doi:10.1093/annonc/mdg019\\u003c/li\\u003e\\n \\u003cli\\u003eBaar MP, Brandt RM, Putavet DA, et al. Targeted apoptosis of senescent cells. \\u003cem\\u003eCell\\u003c/em\\u003e. 2017;169(1):132-47. Doi:10.1016/j.cell.2017.02.031\\u003c/li\\u003e\\n \\u003cli\\u003eMerz LE, Story CM, Osei MA, et al. Absolute neutrophil count by Duffy status. \\u003cem\\u003eBlood Adv\\u003c/em\\u003e. 2023;7(3):317-20. Doi:10.1182/bloodadvances.2022007754\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"},{\"header\":\"Tables\",\"content\":\"\\u003cp\\u003eTables 1 to 3 are available in the Supplementary Files section.\\u003c/p\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":true,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"bmc-cancer\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"bcan\",\"sideBox\":\"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)\",\"snPcode\":\"\",\"submissionUrl\":\"https://www.editorialmanager.com/bcan/default.aspx\",\"title\":\"BMC Cancer\",\"twitterHandle\":\"BMC_series\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"em\",\"reportingPortfolio\":\"BMC Series\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Colorectal cancer, Chemotoxicity, Artificial Intellignece, Machine learning, Prediction, Risk Factors, Biological aging, Social determinants of health, Geospatial variations\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-6628340/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-6628340/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003e\\u003cstrong\\u003eBackground:\\u003c/strong\\u003e Colorectal cancer patients often face chemotoxicity, impacting treatment adherence, survival, and quality of life. Early chemotoxicity screening is vital, yet comprehensive predictive models are lacking. We aimed to develop artificial intelligence(AI)/machine learning (ML)-based models to predict global, gastrointestinal (GI), and hematological chemotoxicity by incorporating racialized group, social determinants of health (SDOH, including Area Deprivation Index measuring geospatial variation) and biological aging (measured by blood-based Levine PhenoAge).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eMethods:\\u003c/strong\\u003e We used electronic health records data from 1,735 adult CRC patients. Sociodemographic/clinical variables, Levine PhenoAge (biological aging), and SDOH (including geospatial data measured by Area Deprivation Index) were analyzed using descriptive statistics. Associations with chemotoxicity (global, GI, hematological) were evaluated via univariate tests. Significant predictors from univariate tests were selected for AI/ML modeling. Six supervised ML models were trained on 80% of cases (n=1,388), with 20% (n=347) reserved for testing. Performance was assessed via accuracy, area under the curve (AUC), and F1-score. Permutation feature importance ranked predictors to define the most significant predictors of chemotoxicity.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eResults:\\u003c/strong\\u003e Chemotoxicity incidences over 6 months of chemotherapy were 56% (global), 41% (GI), and 23% (hematological). Support Vector Machine, followed by XGBoost models (in both training and test datasets) demonstrated high accuracy. Key predictors for global and GI toxicities included advanced biological aging (higher Levine PhenoAge), elevated inflammatory markers (e.g., C-reactive protein), and poor SDOH including geospatial variations (e.g., higher Area Deprivation Index), unemployment. Hematological toxicity was linked to lower immune markers and higher biological age (Levine PhenoAge). Race (non-Hispanic Black), body mass index, and lifestyles also influenced global and GI toxicities.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eConclusions\\u003c/strong\\u003e: The ML models demonstrated high accuracy in chemotoxicity prediction. Biological aging and SDOH, including ADI, and immune/inflammation markers, were common risk factors of global and GI chemotoxicities. In contrast, biological age and immune/inflammation markers were only linked to hematological chemotoxicity. Integrating these factors into predictive models can help clinicians identify at-risk patients and tailor interventions (e.g., anti-inflammatory and anti-aging strategies) to reduce chemotoxicity and improve survivorship outcomes.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Machine Learning-based Chemotoxicity Predictions in Patients with Colorectal Cancer: Integrating Race, Geospatial Social Determinants of Health, and Biological Aging\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-05-19 05:33:20\",\"doi\":\"10.21203/rs.3.rs-6628340/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"decision\",\"content\":\"Revision requested\",\"date\":\"2025-07-01T11:51:15+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-06-18T19:30:22+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"298420697406815188869135928542521631582\",\"date\":\"2025-06-16T13:52:45+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"330215620431378278977019031982424221282\",\"date\":\"2025-06-14T17:53:18+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-06-04T16:36:17+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"90928337102021945326400139838688419237\",\"date\":\"2025-05-23T17:57:43+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"317662091708906007215807040346152575229\",\"date\":\"2025-05-23T11:59:26+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"130704443238656417254231360638731612801\",\"date\":\"2025-05-23T07:24:56+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewersInvited\",\"content\":\"\",\"date\":\"2025-05-23T05:23:30+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2025-05-22T15:37:21+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvited\",\"content\":\"\",\"date\":\"2025-05-13T20:04:15+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2025-05-13T19:25:52+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"BMC Cancer\",\"date\":\"2025-05-13T19:24:50+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"bmc-cancer\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"bcan\",\"sideBox\":\"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)\",\"snPcode\":\"\",\"submissionUrl\":\"https://www.editorialmanager.com/bcan/default.aspx\",\"title\":\"BMC Cancer\",\"twitterHandle\":\"BMC_series\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"em\",\"reportingPortfolio\":\"BMC Series\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"8d387703-795d-41b7-89a6-f3a4ad6f2167\",\"owner\":[],\"postedDate\":\"May 19th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"published-in-journal\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2025-10-13T16:09:02+00:00\",\"versionOfRecord\":{\"articleIdentity\":\"rs-6628340\",\"link\":\"https://doi.org/10.1186/s12885-025-14831-4\",\"journal\":{\"identity\":\"bmc-cancer\",\"isVorOnly\":false,\"title\":\"BMC Cancer\"},\"publishedOn\":\"2025-10-06 15:57:01\",\"publishedOnDateReadable\":\"October 6th, 2025\"},\"versionCreatedAt\":\"2025-05-19 05:33:20\",\"video\":\"\",\"vorDoi\":\"10.1186/s12885-025-14831-4\",\"vorDoiUrl\":\"https://doi.org/10.1186/s12885-025-14831-4\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-6628340\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-6628340\",\"identity\":\"rs-6628340\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}