• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Friday, November 20, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Machine Learning Model Predicts Major Depression Risk in Lymphoma Patients

by
November 20, 2026
in Health
Reading Time: 5 mins read
0
Machine Learning Model Predicts Major Depression Risk in Lymphoma Patients

Machine Learning Model Predicts Major Depression Risk in Lymphoma Patients

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

In a retrospective study analyzing clinical records from patients undergoing treatment for diffuse large B-cell lymphoma, researchers developed and internally evaluated machine learning algorithms to predict the new onset of major depressive disorder. Diffuse large B-cell lymphoma represents the most prevalent subtype of non-Hodgkin lymphoma, and while traditional immunochemotherapy has improved overall survival rates, severe psychiatric comorbidities such as major depressive disorder frequently complicate the clinical course. Depression in this patient population is correlated with worse survival outcomes, longer hospitalizations, and decreased adherence to antineoplastic therapy. Identifying patients who are likely to develop depressive episodes during early treatment stages could enable timely psychological support, yet the complex onset of the condition has historically made early identification difficult in routine oncology practice.

To investigate whether baseline clinical data could forecast psychiatric vulnerability, the research team analyzed a retrospective cohort of 526 patients with pathologically confirmed diffuse large B-cell lymphoma treated between January 2018 and December 2025 at the Department of Hematology and Oncology at Leshan Hospital of Traditional Chinese Medicine. All included patients were at least 18 years of age and scheduled to receive standard R-CHOP immunochemotherapy consisting of rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone. The investigators excluded individuals who had documented histories of major psychiatric disorders prior to lymphoma diagnosis, secondary primary malignancies, severe acute infections, unstable autoimmune conditions, or incomplete laboratory and psychological records at baseline. The primary clinical outcome was the new onset of major depressive disorder during chemotherapy, formally diagnosed by certified psychiatrists according to criteria from the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition.

To overcome potential algorithm bias arising from severe class imbalance in standard observational settings, the researchers implemented an intentional retrospective enrichment strategy at the data extraction stage by oversampling records with documented mood fluctuations or psychiatric consultations. This resulted in an analytic cohort comprising 247 patients who developed new-onset major depressive disorder and 279 patients who did not, reflecting an approximate 1:1 balance. In the total cohort of 526 patients, baseline demographic characteristics were closely comparable between groups, with an overall mean age of 57.4 years and a mean body mass index of 23.6 kilograms per square meter. Neither age nor body mass index differed significantly between patients who later developed depression and those who remained depression-free.

A total of fifteen baseline features were extracted prior to the initiation of chemotherapy, encompassing routine hematological, biochemical, endocrine, psychological, and oncological staging parameters. Laboratory indicators included hemoglobin, albumin, beta-2 microglobulin, lactate dehydrogenase, C-reactive protein, neutrophil-to-lymphocyte ratio, and interleukin-6. Endocrine stress and baseline psychological status were captured using morning serum cortisol measurements and the Patient Health Questionnaire-9 score, while tumor burden and performance status were documented through the Ann Arbor stage, International Prognostic Index score, Eastern Cooperative Oncology Group performance status, and the presence of constitutional B symptoms such as fever, night sweats, and weight loss.

Patients who subsequently developed major depressive disorder exhibited distinct physiological and clinical baselines compared to those who did not. While median baseline Patient Health Questionnaire-9 scores were statistically different between groups—registering a median of 5 in the depression cohort versus 4 in the non-depression cohort—both values fell well within the subclinical, sub-threshold range, indicating that patients were not clinically depressed prior to initiating chemotherapy. However, patients developing depression presented with significantly higher baseline levels of systemic inflammatory and endocrine markers, including a median interleukin-6 of 28.5 picograms per milliliter compared to 16.4 picograms per milliliter, and a mean morning serum cortisol of 472.4 nanomoles per liter compared to 405.4 nanomoles per liter. Furthermore, advanced Ann Arbor stages of III to IV were present in 78.5 percent of the depression group versus 51.3 percent of the non-depression group, and constitutional B symptoms were present in 70.0 percent compared to 29.7 percent.

The researchers split the dataset into an 80 percent training set containing 421 patients and a 20 percent independent hold-out testing set containing 105 patients to build and evaluate six distinct machine learning algorithms. Hyperparameter optimization was conducted strictly within the training set using ten-fold stratified cross-validation alongside grid search techniques to prevent data leakage and avoid overfitting. Among the six evaluated models—which included Random Forest, eXtreme Gradient Boosting, Gradient Boosting Machine, Support Vector Machine, K-Nearest Neighbors, and multivariable Logistic Regression—the ensemble tree-based Random Forest algorithm demonstrated the strongest overall discriminative performance. It achieved a mean cross-validated area under the receiver operating characteristic curve of 0.908 in training and an area under the curve of 0.913 in the independent testing set, outperforming the baseline logistic regression model which attained an area under the curve of 0.756.

In the independent testing set, the Random Forest model maintained balanced diagnostic accuracy across multiple evaluation metrics when assessed at the optimal Youden index threshold. The model yielded an overall accuracy of 0.838, a sensitivity of 0.846, a specificity of 0.830, and an F1-score of 0.838, based on a confusion matrix reflecting 44 true positives, 44 true negatives, 9 false positives, and 8 false negatives. Precision-recall analysis demonstrated an area under the curve of 0.906, helping assess the potential to limit unnecessary psychiatric referrals. Beyond discriminatory capacity, probability calibration proved robust, with the Random Forest achieving a Brier score of 0.124, a calibration intercept of minus 0.018, and a calibration slope of 0.964, alongside a non-significant Hosmer-Lemeshow test confirming that predicted probabilities aligned closely with observed event rates.

To provide model interpretability and assess how individual variables drove mathematical risk calculations, the authors applied SHapley Additive exPlanations, known as SHAP. Global feature importance rankings established that constitutional B symptoms served as the most influential predictive variable, followed in order by interleukin-6, baseline Patient Health Questionnaire-9 score, lactate dehydrogenase, and morning serum cortisol. Recursive feature elimination confirmed that model performance peaked when utilizing these five leading predictors and plateaued thereafter. Clinically, constitutional B symptoms represent substantial physical disease burden and systemic cytokine release, while baseline questionnaire scores capture subtle pre-treatment affective vulnerability that can predispose individuals to full depressive episodes under the physical strain of cytotoxic chemotherapy.

Local SHAP dependence plots and Individual Conditional Expectation curves demonstrated non-linear predictive associations across primary continuous biomarkers. Model-predicted risk rose sharply beyond approximate inflection points of 25 picograms per milliliter for interleukin-6 and 480 nanomoles per liter for morning serum cortisol. Beyond these thresholds, marginal feature contributions shifted from near-zero to positive values. The authors noted that age exhibited a U-shaped association, indicating higher model-estimated probabilities at both younger and older extremes relative to age 45, while baseline psychological scores displayed a direct positive linear trajectory. Multidimensional interaction analyses further revealed that concurrent elevations in interleukin-6 and cortisol were statistically linked with disproportionately higher model-predicted risks compared to elevations in either biological marker alone.

The study authors emphasized several critical limitations inherent in their findings. Because the investigation was retrospective and conducted at a single medical center, external validity remains unverified across broader populations and diverse health systems. Furthermore, the enriched 1:1 case-control design produced an artificial depression prevalence of 46.9 percent, which exceeds unselected clinical rates and mandates formal mathematical recalibration prior to real-world application. The researchers also cautioned that SHAP values reflect statistical associations within a mathematical model rather than verified biological causal mechanisms or validated clinical decision thresholds. Additionally, routine measurement of interleukin-6 and morning cortisol requires standardized collection protocols to account for diurnal variation and introduces testing costs that warrant dedicated health-economic evaluation before such screening panels can be integrated into routine oncology care.

Subject of Research: Machine learning prediction of new-onset major depressive disorder in diffuse large B-cell lymphoma

Article Title: Predicting new-onset major depressive disorder in diffuse large B-cell lymphoma via machine learning: insights into neuro-immune interactions using SHAP

Article References: Yan, K., Chen, L., Liao, X., Han, Y., Zhou, K., Song, Y., Tang, L., & Li, Y. (2026). Predicting new-onset major depressive disorder in diffuse large B-cell lymphoma via machine learning: insights into neuro-immune interactions using SHAP. Frontiers in Psychiatry, 17, Article 1868471. https://doi.org/10.3389/fpsyt.2026.1868471

Image Credits: AI Generated

DOI: 10.3389/fpsyt.2026.1868471

Keywords: diffuse large B-cell lymphoma, major depressive disorder, machine learning, Random Forest, SHAP analysis, neuro-immune axis, interleukin-6, cortisol, lactate dehydrogenase, B symptoms, PHQ-9, oncology, psychiatric oncology, cancer comorbidities, computational psychiatry, biomarkers

News Source: Glenn Wilkins. (November 20, 2026). Machine Learning Model Predicts Major Depression Risk in Lymphoma Patients. Scienmag.

Tags: B symptomsbiomarkerscancer comorbiditiesComputational Psychiatrycortisoldiffuse large B-cell lymphomainterleukin-6lactate dehydrogenaseMachine Learningmajor depressive disorderneuro-immune axisOncologyPHQ-9psychiatric oncologyRandom ForestSHAP analysis
Share12Tweet7Share2ShareShareShare1

Related Posts

Knee Alignment Analysis Reveals Sex Differences in Chinese Osteoarthritis Patients

Knee Alignment Analysis Reveals Sex Differences in Chinese Osteoarthritis Patients

November 20, 2026
Restoring Sinus Rhythm Before Catheter Ablation Lowers Persistent Atrial Fibrillation Recurrence

Restoring Sinus Rhythm Before Catheter Ablation Lowers Persistent Atrial Fibrillation Recurrence

November 20, 2026

Endoscopic Double Bone Channel Technique Shows Biomechanical Stability in Cervical Simulation

November 20, 2026

Faricimab Injections Improve Deep Choroidal Blood Flow in Diabetic Macular Edema

November 20, 2026

POPULAR NEWS

  • CMOS Potts Chip Harnesses Multiphase Oscillations for Low-Power Memory and Graph Optimization

    CMOS Potts Chip Harnesses Multiphase Oscillations for Low-Power Memory and Graph Optimization

    29 shares
    Share 12 Tweet 7
  • Optimized Radiolabeling Improves iPSMA Homogeneity and Supports Single Vial Cold Kit Formulation

    29 shares
    Share 12 Tweet 7
  • Cross-Sectional Bahrain Survey Links Weather, Demands, and Gender to Activity Rates

    29 shares
    Share 12 Tweet 7
  • Biologists Propose Teleonome Framework to Assess Animal Welfare and Agency

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Access Barriers and Institutional Inclusion Shape University Platform Usability in Survey

Obstacle-Node Algorithm Improves Robotic Arm Path Planning for Catenary Maintenance

Knee Alignment Analysis Reveals Sex Differences in Chinese Osteoarthritis Patients

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.