• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Monday, November 16, 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 Biology

Refined Time Intervals Improve Survival Model Predictions for Shifting COVID-19 Risks

by
November 9, 2026
in Biology, Technology
Reading Time: 6 mins read
0
Refined Time Intervals Improve Survival Model Predictions for Shifting COVID-19 Risks

Refined Time Intervals Improve Survival Model Predictions for Shifting COVID-19 Risks

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Dividing follow-up time into refined intervals and applying a weighted aggregation approach significantly sharpens the predictive accuracy of survival models while exposing dramatic shifts in mortality risks across emerging SARS-CoV-2 variants, according to a study published in Genomics and Informatics. Evaluating large-scale observational data from the UK Biobank and synthetic datasets, researchers established that stratified Cox proportional hazards frameworks overcome estimation hurdles when tracking infectious diseases across shifting epidemiological phases. Expanding temporal divisions from five broad time blocks into fifteen calibrated intervals replaced a previously pooled pre-Delta hazard ratio of 7.333 with distinct, variant-specific estimates: 29.359 for the Early variant, 20.734 for the EU1 variant, and 4.079 for Alpha. This demonstrated a marked reduction in death risk over sequential viral lineages. The findings illustrate that when the proportional hazards assumption is satisfied through interval segmentation, standard statistical frameworks maintain superior ranking ability over machine learning and deep learning survival models.

The analytical demand for robust survival modeling became apparent during the COVID-19 pandemic, where quantifying patient risk factors and survival duration proved critical for informing clinical disease management and treatment strategies. Survival analyses have informed research across diverse viral pathogens, including Ebola virus and human metapneumovirus, yet respiratory epidemics present modeling complexities that challenge conventional tools. Analysts generally approach time-to-event outcomes in infectious disease cohorts through two distinct frameworks: measuring elapsed survival from initial infection to death or last follow-up, or calculating survival from the start of the study period onward. While the infection-based timeline provides an intuitive assessment of how environmental variables alter survival, the study-start approach introduces mathematical challenges because infections emerge continuously across the timeline and their clinical effects typically decline over follow-up. This reality generates two distinct temporal forces: infection hazard ratios shift across periods, representing time-dependent coefficients, while individual patient infection status changes dynamically from uninfected to infected, representing time-dependent covariates.

Standard Cox proportional hazards models rely on an assumption of constant covariate effects over time, leaving them ill-equipped to accommodate the rapidly evolving dynamics and fluctuating hazard ratios of infectious disease outbreaks. In prior work, investigators attempted to address non-proportionality by deploying stratified Cox models partitioned into five intervals defined by population surveillance data indicating when specific variants exceeded ninety-five percent prevalence. While that earlier investigation confirmed that mortality hazards peaked during the Early variant stage and dropped lowest during Omicron waves, it treated the extensive duration preceding Delta as one single, uniform interval. That broad grouping satisfied mathematical proportionality criteria but prevented researchers from isolating risk discrepancies between the earliest historical strains and the subsequent Alpha phase. Furthermore, earlier investigations did not benchmark their stratified Cox models against modern statistical learning frameworks, leaving unresolved questions regarding whether flexible machine learning algorithms or neural networks could yield superior survival predictions under time-dependent risks.

To establish empirical clarity, the research team examined a curated cohort derived from the UK Biobank, which recruited 503,325 volunteers aged forty to sixty-nine across twenty-two research centers between 2006 and 2010. The investigators selected 430,747 individuals of European ancestry possessing high-quality genotype data and clinical records updated through December 19, 2022. Within this cohort, records documented 100,827 verified SARS-CoV-2 cases and 329,920 controls, with infected participants classified across seven variant groups: Early variant, EU1, Alpha, Delta, Omicron, Omicron2, and Omicron3. Stringent genomic filtering retained 9,572,556 single nucleotide polymorphisms by excluding markers deviating from Hardy-Weinberg equilibrium, displaying call or sample missingness over ten percent, showing minor allele frequencies below 0.01, or exhibiting imputation scores below 0.8. Using genome-wide association summary statistics generated via BOLT-LMM across a ten-fold cross-validation scheme, the authors calculated standardized polygenic risk scores for type 2 diabetes with LDpred, choosing causal variant tuning parameters that maximized the area under the curve.

The investigative team systematically benchmarked four survival methodologies to determine how parametric and algorithmic models handle evolving clinical covariates: the stratified Cox proportional hazards model, random survival forest, DeepSurv, and DeepHit. The stratified Cox model serves as a traditional parametric instrument allowing each time interval to have a distinct reference hazard function while presuming proportional hazards within intervals. In contrast, the random survival forest functions as a nonparametric machine learning architecture constructing an ensemble of survival trees from bootstrap samples to estimate cumulative hazard functions. DeepSurv deploys a deep feed-forward neural network to parameterize individual hazard functions, utilizing scaled exponential linear units, adaptive moment estimation, learning rate scheduling, and L2 regularization to prevent overfitting while minimizing average negative log partial likelihood. DeepHit employs a deep neural network suited for competing risks, combining the log-likelihood of joint first hitting times and events with cause-specific ranking loss functions to optimize predictive discrimination.

Model performance across both synthetic benchmarks and observational datasets was quantified using the time-dependent concordance index, which measures a model’s ability to rank survival probabilities across dynamic time points. In simulated environments satisfying proportional hazards, survival times were generated via a Weibull baseline hazard function incorporating shape and scale parameters alongside clinical covariates including age, sex, body mass index, type 2 diabetes status, polygenic risk scores, genotyping arrays, and variants. Evaluating one hundred simulation iterations across event rates of five and ten percent with interval censoring rates of zero and five percent confirmed that the Cox model consistently achieved the highest concordance index. DeepSurv and DeepHit demonstrated comparable accuracy, outperforming random survival forests but failing to surpass the Cox model. Applying five percent interval censoring slightly lifted concordance under five percent event rates by reducing comparable pairs, whereas under ten percent event rates, interval censoring reduced predictive precision through event loss.

When simulations introduced time-dependent covariate dynamics via an interaction between the SARS-CoV-2 infection indicator and log-transformed time, the analytical necessity of interval partitioning became evident. Fitting an unstratified Cox model without interval division resulted in immediate violations of the proportional hazards assumption, precluding valid evaluation. Segmenting survival timelines into five intervals of 200 days or ten intervals of 100 days using survSplit restored mathematical compliance and boosted predictive accuracy. The Cox model retained the highest time-dependent concordance scores across evaluated conditions, followed by DeepHit, DeepSurv, and random survival forests. Simulation trials revealed that dropping event rates from ten to five percent decreased concordance and expanded variance, while omitting interval censoring yielded superior performance compared to five percent interval censoring. The most pronounced performance decline occurred when simulated infection rates dropped from fifty to twenty-five percent, underscoring that depriving models of data on the most influential predictor heavily degrades predictive discernment.

Translating these methodological insights into real-world observational data, the authors analyzed all 430,747 UK Biobank participants to model mortality risk from baseline study entry through death or final follow-up. The legacy five-interval model established boundaries at day cutoffs of 494, 672, 757, and 851, corresponding to surveillance data when successive viral variants reached dominant ninety-five percent population prevalence. To ensure strict adherence to proportional hazards criteria, the investigators evaluated each interval using the Schoenfeld residuals test, iteratively dividing any violating segment into two equal-length sub-intervals until all segments satisfied proportionality. This recursive calibration yielded a fifteen-interval model defined by fourteen cutoffs: 111, 166, 194, 222, 284, 346, 364, 383, 402, 420, 494, 672, 757, and 851 days. Ten-fold cross-validation demonstrated that expanding the time-interval framework from five to fifteen segments substantially elevated the average concordance index for both stratified Cox models and DeepHit, confirming that finer segmentation empowers statistical architectures to capture granular shifts in mortality.

To synthesize the fifteen interval-specific hazard estimates into coherent summaries for individual viral strains, the researchers introduced a weighted sum framework where each interval’s estimated effect is scaled by its duration relative to the total lifespan of the corresponding variant. Standard analytical formulas for independent estimates provided final aggregated hazard ratios and standard errors across all seven recognized COVID-19 variants, revealing that the Early variant carried an integrated hazard ratio of 29.359, which declined to 20.734 for EU1 and dropped to 4.079 for Alpha. When the investigators performed survival analysis restricted solely to 100,827 infected individuals measuring survival from infection date, the proportional hazards assumption was violated. In this subgroup analysis, machine learning and deep learning models including random survival forests, DeepSurv, and DeepHit outperformed the traditional Cox model. Incorporating variant classifications alongside clinical and genetic covariates improved predictive capacity across models, confirming that viral strains inherently modify patient hazard levels.

Despite these notable computational advancements, the authors emphasized critical methodological limitations that warrant consideration in future infectious disease modeling. The current practice of bisecting non-proportional intervals into equal-length segments presents operational risks, as mechanical partitioning can generate narrow windows containing very few mortality events, thereby compromising numerical stability. Developing adaptive partitioning strategies that ensure balanced event distributions across intervals represents an essential next step for refining time-dependent parameter estimation. Additionally, while the weighted sum technique provides an intuitive means of aggregating fragmented interval coefficients into variant-level hazard ratios, the authors noted that its capacity to fully represent complex underlying time-varying effects requires dedicated simulation validation. Future empirical studies must also conduct expanded simulation trials to determine whether deep learning and machine learning algorithms consistently outperform proportional hazards models whenever non-proportionality occurs in observational cohorts. Advancing these interval-based survival techniques will prove vital for enhancing surveillance and predictive precision across evolving pathogen landscapes.

Subject of Research: Statistical and machine learning survival modeling for time-dependent infectious disease outcomes

Article Title: Statistical learning methods for improving predictive performance in time-dependent survival models

Article References: Seo, H., & Chung, W. (2025). Statistical learning methods for improving predictive performance in time-dependent survival models. Genomics & Informatics, 23(1), Article 19. https://doi.org/10.1186/s44342-025-00050-7

Image Credits: AI Generated

DOI: 10.1186/s44342-025-00050-7

Keywords: survival analysis, Cox proportional hazards, machine learning, deep learning, COVID-19 variants, time-dependent covariates, UK Biobank, predictive performance, Statistical, learning, methods, improving

News Source: Blake Davidson. (November 9, 2026). Refined Time Intervals Improve Survival Model Predictions for Shifting COVID-19 Risks. Scienmag.

Tags: COVID-19 variantsCox proportional hazardsdeep learningImprovinglearningMachine Learningmethodspredictive performancestatisticalsurvival analysistime-dependent covariatesUK Biobank
Share12Tweet7Share2ShareShareShare1

Related Posts

Engineered Male Sterility Systems Advance Scalable Hybrid Seed Production in Maize

Engineered Male Sterility Systems Advance Scalable Hybrid Seed Production in Maize

November 16, 2026
YOD1 Stabilization Promotes Bcl-xL-Dependent Enzalutamide Resistance in Prostate Cancer Models

YOD1 Stabilization Promotes Bcl-xL-Dependent Enzalutamide Resistance in Prostate Cancer Models

November 16, 2026

Body Region Aware Framework Achieves State-of-the-Art Occluded Person Re-Identification

November 16, 2026

Subsurface Models Identify Groundwater Archetypes for Geothermal Energy Across Berlin

November 16, 2026

POPULAR NEWS

  • Childhood Plasma miRNAs Indicate Neurodevelopmental Delay Linked to Prenatal Alcohol Exposure

    Childhood Plasma miRNAs Indicate Neurodevelopmental Delay Linked to Prenatal Alcohol Exposure

    29 shares
    Share 12 Tweet 7
  • OCD Patients Show Heightened Sensitivity to Local Evidence Despite Lower Overall Confidence

    29 shares
    Share 12 Tweet 7
  • Prenatal Anxiety and Depression Predict Distinct Infant Cerebellar Asymmetry Patterns

    29 shares
    Share 12 Tweet 7
  • Probing combined with platelet-rich plasma improved moderate-to-severe meibomian dysfunction in trial

    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

Lifelong Serotonin Deficiency Delays Maturation and Alters Sociability in Rats

AI Guidance Enables Nonexperts to Capture Diagnostic Echocardiograms in Controlled Settings

Prenatal Stress Shows No Link to Epigenetic Age Acceleration at Birth

Subscribe to Blog via Email

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

Join 88 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.