The research team, led by Rasheed Omobolaji Alabi of the University of Helsinki together with collaborators at the University of Vaasa, Uppsala University, and Karolinska Institute, drew on 1,525 patient records from the Surveillance, Epidemiology, and End Results (SEER) Program, the National Cancer Institute’s vast population-based registry covering a substantial fraction of the United States. All included patients had histologically confirmed squamous cell carcinoma arising in the lip and oral cavity, oropharynx, hypopharynx, or larynx, had not undergone surgery, and had received either definitive radiotherapy alone or concurrent chemoradiotherapy. The cohort was overwhelmingly male, with 1,221 men and 304 women, and the mean age at diagnosis was 73.1 years, ranging up to 90.

What distinguishes this analysis from previous comparisons of the two treatment modalities is its methodological architecture. The researchers combined classical survival statistics with machine learning techniques in a way that had not previously been applied to this specific clinical question. Kaplan-Meier curves and multivariable Cox proportional hazards regression provided the traditional backbone, while an extreme gradient boosting model, trained alongside logistic regression and easy ensemble algorithms in Azure Machine Learning Studio and Jupyter notebooks, was used to probe how each patient characteristic contributed to disease-specific survival. The model was validated with five-fold cross-validation, a technique that repeatedly splits the data into training and testing subsets to guard against overfitting in the absence of an external validation cohort.

The survival results were unambiguous. Kaplan-Meier curves for the two treatment groups separated early and remained distinct throughout follow-up, which ranged from zero to 167 months, with the log-rank test yielding a p-value below 0.001. In the Cox regression, chemoradiotherapy carried a hazard ratio of 0.699, with a 95 percent confidence interval of 0.598 to 0.818, making it the strongest protective factor in the entire model. Conversely, the presence of distant metastasis emerged as the most powerful adverse prognostic factor, with a hazard ratio of 2.520, while hypopharyngeal tumor site and increasing age also independently predicted worse disease-specific survival.

The machine learning layer of the study added a different kind of insight. The extreme gradient boosting model achieved a performance accuracy of 55.5 percent in predicting disease-specific survival, with an area under the receiver operating characteristics curve of 0.57, figures the authors themselves acknowledge as modest. But predictive accuracy was never the primary goal. Instead, the team deployed permutation feature importance, a model-agnostic ranking method that works by randomly shuffling each input variable one at a time and measuring how much the model’s performance degrades. Features whose shuffling causes the largest drop in accuracy are deemed the most influential. In this analysis, the four most important features for survival were age, metastatic stage, overall TNM stage, and, critically, the treatment option itself.

To make the model’s inner workings transparent, the researchers also applied SHapley Additive exPlanations, or SHAP, a technique borrowed from cooperative game theory that assigns each predictor a quantified contribution to individual predictions. The SHAP beeswarm plot confirmed the global picture: higher age, advanced stage, and distant metastasis pushed predicted mortality risk upward, while treatment with chemoradiotherapy contributed substantially toward improved survival predictions, and being married was also associated with better outcomes. A local SHAP waterfall plot for a single illustrative patient showed how treatment status and tumor grade were the strongest contributors to reducing that individual’s predicted risk, demonstrating the kind of patient-specific interpretability that clinicians increasingly demand from artificial intelligence tools.

The propensity-score matching analysis added an important layer of caution. Using logistic regression to estimate each patient’s propensity for receiving chemoradiotherapy, and then matching patients one-to-one by nearest-neighbor matching without replacement, the team sought to reduce the treatment-selection bias that plagues any non-randomized comparison. The analysis revealed substantial differences in nearly all covariates between the two treatment groups, with patients receiving chemoradiotherapy generally exhibiting higher propensity scores and more concentrated distributions. In other words, healthier or differently profiled patients were systematically more likely to receive the combined treatment, a reality that means the observed survival advantage, while consistent across every analytical method, cannot be interpreted as definitively causal.

The clinical context makes these findings consequential. Head and neck cancer is among the ten most common malignancies worldwide, with an estimated 54,000 new cases and about 11,230 deaths in the United States in 2022, and between 25 and 40 percent of patients are diagnosed at age 70 or older, most already at an advanced stage. Yet older patients remain underrepresented in clinical trials, leaving clinicians with a dearth of evidence about whether they can tolerate aggressive treatment. Guidelines from the National Comprehensive Cancer Network, the American Society of Clinical Oncology, and the European Society for Medical Oncology recommend concurrent chemoradiotherapy for advanced disease, but the evidence base for older patients specifically has been thin, and conflicting reports have appeared, with some studies favoring radiotherapy alone. The new machine learning analysis aligns with the randomized trial literature suggesting that chemoradiotherapy yields better disease-specific survival, and it echoes the authors’ argument that treatment decisions should rest on patient-related factors rather than chronological age alone.

The study also situates itself within a broader technological transformation of head and neck oncology. The authors discuss how FDG-PET/CT-based radiotherapy planning can refine staging and enable biologically informed dose escalation to metabolically active tumor regions, how Internet of Things devices enable continuous remote monitoring of microvascular free-flap reconstructions after surgery, and how three-dimensionally printed patient-specific surgical plates have measurably improved the accuracy of bony reconstruction. These developments, they argue, illustrate how data-driven and digitally enabled medicine is reshaping every stage of cancer care, from imaging and treatment planning to postoperative surveillance.

The limitations of the study are substantial and the authors are candid about them. The SEER database lacks information on HPV status, a key prognostic factor in oropharyngeal cancer; on comorbidities and frailty, both of which strongly influence survival and treatment tolerance in older patients; on the specific chemotherapy regimen, its dose, and its completeness; and on the type and intent of radiotherapy. The retrospective design introduces selection bias, and the uneven distribution of tumor sites, dominated by laryngeal and oral cavity cancers, may skew the analysis. The modest accuracy of the machine learning model means it cannot yet be used to predict survival for new patients under treatment consideration, and the authors emphasize that the findings should be regarded as exploratory. Still, the convergence of Kaplan-Meier curves, Cox regression, permutation feature importance, and SHAP analysis on the same conclusion gives the result unusual internal consistency. The team calls for multicenter randomized controlled trials and age-specific prospective studies, with separate analyses for different tumor sites, to confirm the benefit. Until then, the study offers older patients and their physicians something they have rarely had: large-scale, algorithmically interrogated evidence that age alone should not disqualify anyone from the most effective nonsurgical treatment available.

Subject of Research: Machine learning comparison of chemoradiotherapy versus radiotherapy alone for disease-specific survival in older patients with advanced-stage, non-surgically treated head and neck squamous cell carcinoma

Article Title: Effectiveness of radiotherapy or chemoradiotherapy in older patients with advanced-stage, non-surgically treated head and neck cancer: A machine learning analysis

Article References: Alabi, R. O., Almangush, A., Elmusrati, M., Tiblom Ehrsson, Y., Laurell, G., & Mäkitie, A. A. (2026). Effectiveness of radiotherapy or chemoradiotherapy in older patients with advanced-stage, non-surgically treated head and neck cancer: A machine learning analysis. Medical Oncology, 43(11), Article 315. https://doi.org/10.1007/s12032-026-03420-5

Image Credits: AI Generated

DOI: 10.1007/s12032-026-03420-5

Keywords: head and neck cancer, chemoradiotherapy, radiotherapy, machine learning, SEER database, disease-specific survival, elderly patients, propensity score matching, SHAP, permutation feature importance, Cox proportional hazards, HPV status

The research team, led by investigators analyzing SEER registry data from 2016 to 2021, identified 11,833 women aged 40 and older with microscopically confirmed, malignant primary breast cancer that had already metastasized to bone at diagnosis. The cohort spanned five racial and ethnic groups: non-Hispanic White, non-Hispanic Black, Hispanic, non-Hispanic Asian or Pacific Islander, and non-Hispanic American Indian or Alaska Native. Every patient had known estrogen receptor, progesterone receptor, and HER2 status, known annual household income, and documented history of surgery, chemotherapy, and radiation. After excluding patients with unknown race or male sex, the researchers followed the cohort for cause-specific survival, meaning deaths attributed to the cancer itself.

The statistical approach was rigorous. The team used Kaplan-Meier survival curves with log-rank tests to compare survival between groups, then built multivariate Cox proportional hazards regression models to isolate the independent effect of race while controlling for age, income, molecular subtype, primary tumor site, and treatment history. The models were evaluated at three timepoints: six months, twelve months, and five years after diagnosis. With 5,019 deaths observed in the cohort, the ratio of events to model parameters was large enough to avoid overfitting, and the proportional hazards assumption was formally tested for each covariate, holding for race, the primary variable of interest.

The headline finding was stark. Non-Hispanic Black women had a 24 percent higher hazard of death from breast cancer compared with non-Hispanic White women over five years (hazard ratio 1.241, 95 percent confidence interval 1.145 to 1.345, p less than 0.001). This elevated risk persisted at every timepoint examined: Black women faced a 21 percent higher risk of death at twelve months and a 26 percent higher risk at five years compared with all other racial groups combined. Meanwhile, non-Hispanic Asian or Pacific Islander women actually had better survival than White women (hazard ratio 0.865), while Hispanic women showed no significant difference from the White reference group at five years.

Income mattered too, but in a way that complicated a simple access-to-care narrative. Women with annual household incomes below $75,000 had a 10 percent higher hazard of death than wealthier women (hazard ratio 0.900 for the higher income group, p less than 0.001), and this disadvantage appeared at six months, twelve months, and five years. Age also played a role, with women 65 and older facing a 15 percent higher hazard of death. But the most provocative result emerged when the researchers stratified by income: among women earning $75,000 or more, Black women still had significantly worse survival than all other races, with a hazard ratio of 1.256 at five years. The disparity held among both younger Black women (hazard ratio 1.242) and those 65 and older (hazard ratio 1.318) within this affluent group.

In other words, affluence did not erase the racial gap. Even when Black women had household incomes well above the national median and had received comparable surgery, chemotherapy, or radiation, they died of their disease at higher rates. The same pattern appeared in the lower-income stratum, where Black women faced a hazard ratio of 1.184 at five years. This finding challenges the long-standing assumption that socioeconomic disparities in breast cancer outcomes are primarily a story about access to screening, insurance coverage, or treatment. The authors note that countries with universal healthcare, including Canada, show similar racial disparities in breast cancer outcomes, suggesting that access alone cannot explain the gap.

Treatment itself was powerfully protective across the board. Women who underwent surgical resection of the primary tumor had a 45 percent lower hazard of death (hazard ratio 0.554), those who received chemotherapy had a 58 percent lower hazard (hazard ratio 0.421), and radiation was associated with a modest 6 percent reduction (hazard ratio 0.940). These associations were consistent at all three follow-up intervals, though the magnitude of benefit for surgery and chemotherapy diminished over time. Molecular subtype also strongly predicted survival: the worst outcomes were seen in the rare ER-negative, PR-positive, HER2-negative subtype (hazard ratio 4.880) and in triple-negative disease (hazard ratio 4.159), both relative to the common ER-positive, PR-positive, HER2-negative reference group.

The biology of the disparity is partially, but not fully, explanatory. Black women in the cohort had roughly twice the odds of having triple-negative breast cancer compared with White women (odds ratio 2.04, 95 percent confidence interval 1.74 to 2.39), a finding consistent with a large body of literature linking triple-negative and basal-like tumors to higher prevalence among Black women. Notably, prior research has connected this elevated triple-negative risk to structural racism itself: Black women living in US states with greater structural racism, measured across domains such as educational inequity and disenfranchisement, have higher odds of triple-negative diagnoses than Black women in less discriminatory states. Yet even after adjusting for molecular subtype in the multivariate model, race remained an independent predictor of mortality, indicating that tumor biology alone does not account for the survival gap.

The authors are careful to frame race as a social construct rather than a biological determinant. They argue that the observed disparities should be understood within the broader context of structural determinants of health: residential segregation, unequal distribution of high-quality cancer care facilities, and disparities in insurance coverage have all been identified as contributors to racial gaps in breast cancer outcomes independent of clinical factors. The study also carries important limitations. SEER does not capture comorbidities, performance status, the burden of metastatic disease, the presence of visceral metastases, or details of systemic therapy such as endocrine treatment, CDK4/6 inhibitors, or HER2-targeted agents, all of which could introduce residual confounding. Requiring complete data on income, receptor status, and treatment may also have excluded disadvantaged patients with missing records, introducing selection bias.

Still, the scale and consistency of the findings make them hard to dismiss. Breast cancer survival has improved dramatically in recent decades: women with localized disease now have five-year survival rates approaching 99 percent, while metastatic disease carries a five-year survival of roughly 31 percent. Bone metastases in particular drive substantial morbidity through pathological fractures and spinal cord compression. Yet Black women’s survival gains have lagged behind those of White women by as much as two decades, according to prior research cited in the study. This new analysis demonstrates that even within the metastatic-to-bone population, where prognosis is uniformly poor, race and income carve out meaningful differences in who survives. The authors call for future research to identify the specific structural, biological, or unmeasured clinical factors driving the excess mortality among Black women, arguing that the gap cannot be entirely attributed to late-stage presentation, income, or the treatments captured in national registry data.

Subject of Research: Racial and socioeconomic disparities in survival among women with breast cancer metastatic to bone

Article Title: Race and Socioeconomic Status are Predictors of Survival in Patients With Metastatic Bone Disease Secondary to Breast Cancer

Article References: Castan, A., Dias, R., Mysore, A., Herz, D., Gotoff, K., Kadkoy, Y., Beebe, K., Benevenia, J., & Ippolito, J. (2026). Race and Socioeconomic Status are Predictors of Survival in Patients With Metastatic Bone Disease Secondary to Breast Cancer. Cancer Reports, 9(10), Article e70689. https://doi.org/10.1002/cnr2.70689

Image Credits: AI Generated

DOI: 10.1002/cnr2.70689

Keywords: breast cancer, bone metastasis, health disparities, survival analysis, SEER database, race, socioeconomic status, triple-negative breast cancer, Cox regression, Kaplan-Meier, structural racism, oncology

The research team, led by Dina Hauptschein, Paul Kang, and Surabhi Amar of the Creighton University Arizona Health Education Alliance, analyzed records from the Surveillance, Epidemiology, and End Results program, the National Cancer Institute’s network of population-based cancer registries. Their final cohort included 227,411 cases of pancreatic ductal adenocarcinoma, or PDAC, diagnosed between 2004 and 2023 across 21 registries covering a substantial portion of the US population. After excluding cases with unknown age, race, or stage, the investigators calculated age-adjusted incidence rates per 100,000 people and used Joinpoint regression, a statistical technique that fits piecewise log-linear models to identify inflection points in disease trends, to estimate how quickly rates were changing over time.

The headline numbers tell a nuanced story. Over the full study period, the age-adjusted incidence of localized PDAC rose with an average annual percent change of 6.00 percent, while distant-stage disease increased only modestly at 1.11 percent per year and regional-stage disease remained essentially flat. Most remarkably, the localized-stage trend accelerated sharply in the latter half of the study period: between 2015 and 2019, the annual percent change in localized diagnoses reached 14.7 percent. Expressed as a share of all staged diagnoses, localized disease climbed from 10.8 percent of cases in 2004 to 19.1 percent in 2023, while distant-stage disease fell from 73.5 percent to 66.4 percent. That is still a sobering majority, but the direction of travel is unmistakable.

Crucially, the trend toward earlier detection was not confined to any single demographic group. Localized-stage incidence rose significantly and comparably in both men and women, with average annual increases of 5.71 percent and 6.29 percent respectively. When the researchers stratified by race and ethnicity, every group showed a significant upward trend in localized diagnoses. Non-Hispanic White individuals saw piecewise annual increases of 14.8 percent between 2015 and 2019, while Non-Hispanic Black individuals experienced similar acceleration of 14.1 percent per year from 2014 to 2019. Hispanic individuals, Non-Hispanic Asian or Pacific Islander individuals, and Non-Hispanic American Indian/Alaska Native individuals all showed significant sustained increases as well, though the latter group displayed greater variability reflecting its smaller population size.

The study also documented persistent and troubling disparities. Non-Hispanic Black individuals had the highest overall incidence of PDAC at 8.91 per 100,000, compared with 7.37 for Non-Hispanic White individuals and 5.38 for Non-Hispanic Asian or Pacific Islander individuals. In multivariable negative binomial regression models adjusting for age, sex, race, and calendar year, Non-Hispanic Black race was associated with a 31 percent higher rate of localized disease and a 43 percent higher rate of distant-stage disease relative to Non-Hispanic White individuals. Prior research has reported pancreatic cancer mortality up to 32 percent higher among Black Americans, and while differences in smoking, obesity, diabetes, healthcare access, and delayed diagnosis likely contribute, the full explanation remains incompletely understood. The new findings underscore that any progress in early detection must grapple with these inequities.

What explains the dramatic post-2015 acceleration in localized diagnoses? The authors point to a convergence of technological and clinical developments. The sheer volume of diagnostic imaging has grown substantially: between 2013 and 2023, the number of CT scans performed in the United States increased by 19.2 percent and MRI scans by 21.4 percent, according to research cited in the study. At the same time, the quality of cross-sectional imaging has improved to the point where smaller pancreatic abnormalities, which would once have escaped notice, are increasingly picked up incidentally during evaluations for unrelated symptoms. Endoscopic ultrasound has also become a routine tool, capable of resolving small lesions and permitting tissue sampling that was previously impractical.

A third driver is the expansion of surveillance programs for people at elevated hereditary risk. Growing recognition of genetic predisposition syndromes involving BRCA1 and BRCA2 mutations, PALB2, Lynch syndrome, hereditary pancreatitis, and familial pancreatic cancer has led to broader implementation of targeted screening using MRI and endoscopic ultrasound in high-risk individuals. Long-term surveillance studies have shown that such programs can detect neoplastic changes before they progress to incurable disease. Meanwhile, researchers are pursuing biomarker-based detection strategies, including protein markers, autoantibodies, circulating tumor DNA, microRNAs, methylated DNA, and exosome-based assays. For patients with pancreatic cysts, cyst-fluid biomarkers such as CEA and mutations in KRAS and GNAS may help risk-stratify potentially precancerous mucinous lesions, though these tests remain adjunctive and require further validation before widespread clinical use.

Yet the authors are careful not to declare victory. The rise in localized diagnoses does not occur in isolation: overall PDAC incidence has also increased across both sexes and all racial and ethnic groups, raising the possibility that at least part of the stage shift reflects a growing underlying disease burden rather than purely earlier detection. Aging populations and the rising prevalence of obesity and diabetes are well-established contributors; according to the CDC’s Behavioral Risk Factor Surveillance System, the percentage of US adults with obesity and diabetes increased by 23 percent and 21 percent respectively between 2011 and 2023. Projections cited in the study suggest pancreatic cancer may become the second-leading cause of cancer-related death in the United States by 2030. More diagnoses of early-stage disease could therefore mean more cancers being found, not fewer cancers progressing.

There is also a methodological caveat that tempers interpretation. Because the SEER database lacks information on screening pathways, individual risk factors, family history, and tobacco use, the study cannot determine why staging patterns changed. More importantly, cases with unknown stage, about 4.7 percent of the identified cases, were excluded from the analysis. If the completeness of staging information improved over the study period, some of the apparent rise in localized incidence could reflect better stage ascertainment rather than a true biological or clinical shift. Changes in staging definitions and reporting practices over two decades could similarly have influenced the results. The authors acknowledge these limitations explicitly and note that the possibility of ascertainment bias cannot be excluded with the available data.

Even with those caveats, the study represents a meaningful milestone in the long struggle against one of medicine’s most stubborn killers. Five-year survival for PDAC remains between 10 and 14 percent, and most patients are still diagnosed with advanced, symptomatic disease. But the demonstration that localized diagnoses are rising across every demographic group, and accelerating in recent years, provides the first population-scale signal that early detection efforts may be gaining traction. Whether that signal translates into longer lives is the critical unanswered question. The authors call for prospective multicenter investigations and randomized trials of targeted screening and surveillance strategies to determine if the growing proportion of localized diagnoses reflects genuine earlier detection, greater diagnostic intensity, an increasing disease burden, or some combination of all three. For a cancer that has defied decades of effort, even a modest, consistent shift in the stage at diagnosis is a development worth watching closely.

Subject of Research: Temporal trends in stage at diagnosis of pancreatic ductal adenocarcinoma in the United States

Article Title: Have we gotten better at diagnosing PDAC before it is too late?: Temporal trends in stage at diagnosis of pancreatic ductal adenocarcinoma in the United States

Article References: Hauptschein, D., Kang, P., & Amar, S. (2026). Have we gotten better at diagnosing PDAC before it is too late?: Temporal trends in stage at diagnosis of pancreatic ductal adenocarcinoma in the United States. Cancer Causes & Control, 37(10), Article 179. https://doi.org/10.1007/s10552-026-02266-6

Image Credits: AI Generated

DOI: 10.1007/s10552-026-02266-6

Keywords: pancreatic cancer, PDAC, early detection, SEER database, cancer staging, incidence trends, endoscopic ultrasound, cancer screening, health disparities, epidemiology, imaging, surveillance

The research team, led by investigators at Tianjin Union Medical Center and Nankai University in China, assembled a substantial dataset to train and test their algorithms. Drawing on the Surveillance, Epidemiology, and End Results database, the large American cancer registry, they identified 3,511 patients who had undergone surgery for gastrointestinal stromal tumors. That cohort was randomly divided at a ratio of seven to three into a training set, used to teach the models, and an internal validation set, used to check their performance on data they had never seen. To guard against the possibility that the models had merely learned quirks of the American registry, the team added a second, entirely independent test: 219 patients treated at their own medical center in Tianjin, whose outcomes served as an external validation cohort.

Before any modeling began, the researchers confronted a problem that plagues every predictive study in oncology: which of the many variables recorded for each patient actually matter? They applied three separate feature selection methods to sift through the candidate predictors, allowing the data itself to determine which variables earned a place in the final models. Notably, the variables that survived this screening process differed between the two prediction tasks. The factors that best forecast death from any cause were not identical to those that best forecast death specifically from the tumor, a distinction that has real clinical consequences, because many patients with these tumors ultimately die of something other than their cancer.

With the variables chosen, the team built four distinct survival models, each representing a different branch of modern machine learning. The random survival forest, an ensemble method that grows many decision trees adapted to censored survival data, competed against a gradient boosting machine, which builds predictions by sequentially correcting the errors of simple models; CoxBoost, a boosting algorithm built directly on the classical Cox proportional hazards framework; and DeepSurv, a neural network approach that extends the Cox model with deep learning. Each model was evaluated with a battery of quantitative metrics, including Harrell’s concordance index, which measures how well a model ranks patients by risk, along with time-dependent areas under the receiver operating characteristic curve, precision-recall analysis, average precision, and decision curve analysis, which estimates the net clinical benefit of acting on a model’s predictions.

The verdict was clear. The random survival forest delivered the most stable predictive performance across both endpoints and both validation settings. In the internal validation cohort, its concordance index reached 0.735 for overall survival and 0.776 for cancer-specific survival, figures that indicate meaningful discrimination between high- and low-risk patients. The model also produced favorable time-dependent AUC values, meaning its accuracy held up across the follow-up period rather than degrading after the first year or two. When the models were confronted with the external cohort of Chinese patients, the random survival forest again demonstrated reliable discrimination and calibration, the latter meaning that when the model predicted, say, a seventy percent probability of surviving five years, roughly seventy percent of such patients did in fact survive that long. Calibration is often the Achilles heel of machine learning models in medicine, which can rank patients correctly while assigning them wildly inaccurate probabilities, so the external confirmation on a demographically distinct population carries particular weight.

Perhaps the most consequential part of the study is not the raw accuracy but the interpretability. Black-box algorithms have faced justified skepticism in clinical medicine, where a prediction that cannot be explained is a prediction that cannot be trusted, audited or acted upon. The researchers addressed this by applying SHAP, or SHapley Additive exPlanations, a technique borrowed from cooperative game theory that assigns each input variable a quantified contribution to every individual prediction. Rather than asking only whether the model is accurate, SHAP reveals why it made each call, showing which features pushed a given patient’s estimated survival up or down and by how much. This transparency allowed the team to identify the dominant drivers of prognosis and to verify that the model was learning biologically plausible relationships rather than statistical artifacts.

The SHAP analysis produced a striking asymmetry between the two survival endpoints. For overall survival, age emerged as the predominant factor, which makes intuitive sense: a model predicting death from any cause will naturally weight the passage of time and its physiological consequences heavily. For cancer-specific survival, however, the picture shifted decisively toward tumor biology, with tumor size and mitotic rate, the measure of how rapidly cancer cells are dividing under the microscope, dominating the predictions. This divergence is more than a statistical curiosity. It tells clinicians that the question a patient is really asking matters. If the concern is how long the patient will live overall, age and its associated comorbidities loom large. If the concern is whether the tumor itself will prove fatal, the aggressiveness of the tumor, captured by its size and its proliferative activity, is what counts. A single one-size-fits-all risk score would blur this distinction.

Translating the model from a research finding into a usable clinical instrument, the team built an online survival calculator based on the random survival forest. The tool allows a clinician to enter an individual patient’s characteristics and receive personalized, time-specific estimates of overall and cancer-specific survival, replacing the coarse risk categories of traditional staging systems with continuous, individualized probabilities. For a fifty-eight-year-old patient with a six-centimeter tumor and a moderate mitotic count, the calculator can generate a survival curve specific to that profile, giving both surgeon and patient a far more granular picture of the road ahead than any nomogram or risk table currently in routine use. Such tools also have the potential to refine follow-up schedules, concentrating surveillance on patients the model flags as high risk while sparing low-risk patients unnecessary imaging and anxiety.

The study arrives amid a broader movement to bring explainable artificial intelligence into oncology, and its design reflects lessons learned from earlier efforts that stumbled. By combining multiple feature selection methods, testing four competing algorithms, validating externally on a separate population, and interrogating the winning model with SHAP, the researchers assembled a methodological chain of evidence that goes beyond a single impressive accuracy figure. The work was approved by the ethics committee of Tianjin Union Medical Center, and the authors declared no competing interests. The study was supported by the Research Project of the Tianjin Municipal Education Commission.

Limitations remain, as they do in any registry-based study. The SEER database, for all its scale, records a defined set of variables and may not capture every factor that influences outcomes, such as the use of targeted therapies like imatinib, molecular mutations in KIT or PDGFRA, or details of surgical technique. The external validation cohort, while independent, comprised 219 patients from a single center, and broader prospective testing will be needed before such a calculator becomes standard practice. Still, the direction of travel is unmistakable. For a tumor whose behavior has long defied simple classification, a model that can weigh tumor size, mitotic rate, age and other variables together, quantify their contributions for each patient, and deliver an honest, calibrated survival estimate represents a genuine step toward precision follow-up care. The era in which a postoperative gastrointestinal stromal tumor patient’s prognosis could be summarized by a three-tier risk label may be drawing to a close, replaced by algorithms that speak in probabilities, show their work, and fit in a browser window.

Subject of Research: Explainable machine learning models for predicting postoperative survival in gastrointestinal stromal tumor patients

Article Title: Prediction of overall and cancer-specific survival in postoperative patients with gastrointestinal stromal tumors based on explainable models

Article References: Chen, Z., Zhao, J., Wang, N., Zhang, J., Song, P., Tian, W., Luo, N., Zhang, S., & Cao, L. (2026). Prediction of overall and cancer-specific survival in postoperative patients with gastrointestinal stromal tumors based on explainable models. BMC Cancer. https://doi.org/10.1186/s12885-026-17022-x

Image Credits: AI Generated

DOI: 10.1186/s12885-026-17022-x

Keywords: gastrointestinal stromal tumor, machine learning, random survival forest, SHAP, overall survival, cancer-specific survival, SEER database, DeepSurv, prognosis, risk stratification, explainable AI, BMC Cancer

The epidemiology has shifted dramatically in recent decades. Historically, studies reported that anywhere from one in five to one in fifty unilateral breast cancer survivors would go on to develop a contralateral tumor, with risk strongly shaped by age at first diagnosis and genetic predisposition. In BRCA1 mutation carriers, ten-year cumulative incidence has been reported as high as 17 percent, particularly among women diagnosed young. Yet contemporary population-based analyses, including studies of the Surveillance, Epidemiology, and End Results (SEER) database, show a significant decline in the standardized incidence of contralateral breast cancer over the past two decades. The decline is attributed primarily to advances in systemic therapy, improved radiotherapy techniques, and deliberate risk-reduction strategies—above all, the widespread adoption of adjuvant endocrine therapy.

That protective effect is now well quantified. A recent meta-analysis found that adjuvant endocrine therapy reduces contralateral breast cancer risk by more than one-third, with the benefit concentrated in estrogen receptor (ER)-positive disease. The flip side is equally instructive: patients whose original tumors were ER- and progesterone receptor (PR)-negative remain at substantially higher risk, and they tend to develop contralateral tumors that are also hormone receptor-negative, regardless of other contributing factors such as germline mutations. This receptor concordance underscores a central theme of the review—hormone receptor status of the first tumor is one of the most powerful predictors of what happens in the opposite breast, and it should anchor any personalized surveillance plan.

Beneath the epidemiology lies a deeper biological question: is a contralateral tumor truly a new cancer, or is it sometimes a metastasis in disguise? The review details how genomic sequencing has largely supported the de novo hypothesis. Whole-genome and transcriptomic studies reveal significant heterogeneity between first and second tumors, with differences in key oncogenic pathways, hormone receptor expression, and mutational burdens. Most contralateral cancers exhibit molecular profiles distinct from the primary tumor, consistent with an independent neoplastic event. However, accumulating evidence shows that a subset shares clonal origins with the initial cancer, raising the possibility of micrometastatic dissemination rather than a wholly new malignancy. Distinguishing between these scenarios matters clinically, because clonally related disease might warrant escalation of systemic therapy, whereas an independent tumor calls for treatment decisions based on its own characteristics.

Molecular subtype adds another layer of stratification. Patients with triple-negative breast cancer face both a higher risk of a contralateral event and a greater chance of an aggressive second phenotype, likely reflecting the younger age at diagnosis and stronger genetic predisposition typical of this subtype. By contrast, women with hormone receptor-positive tumors who receive endocrine therapy experience markedly reduced contralateral incidence, reinforcing the protective role of systemic intervention. Emerging research also implicates the tumor microenvironment—stromal interactions, immune cell infiltration, and epigenetic modifications—in shaping not only whether a contralateral cancer develops but how it behaves and responds to treatment.

Translating these observations into individualized risk estimates remains the field’s central technical challenge. Several prediction models now exist. The Manchester formula, which incorporates age, family history, hormone receptor status, genetic mutations, tumor morphology, and prior treatments, achieves an area under the receiver operating characteristic curve (ROC-AUC) of only 0.61 at five years and 0.60 at ten years—barely better than chance for many clinical purposes. The newer PredictCBC-2.0 model, developed and validated in roughly 200,000 patients, adds parity, nodal status, tumor size, prior chemoradiation, and prior endocrine therapy, reaching ROC-AUC values of 0.66 at five years and 0.65 at ten years; its advantage is applicability to the general, genetically untested survivor population, and an online tool is available. The BRCA-CRisk model, designed for BRCA carriers, performs better still, with ROC-AUC values around 0.78 at five years and 0.70 at ten years. Polygenic risk scores and machine-learning approaches applied to large genomic and clinical datasets may push discrimination further, but the review is candid that current models require refinement before they can reliably drive clinical decisions.

What is not in dispute is the prognostic weight of a contralateral event. Population-based analyses, including SEER data, demonstrate that survivors who develop contralateral breast cancer face a significantly increased risk of breast cancer-specific death compared with those who do not, and this elevated risk persists across subtypes and independently of traditional prognostic factors such as age, stage, and hormone receptor status of the primary tumor. One hypothesis holds that a contralateral cancer is less an isolated event than a marker of an inherent predisposition to tumor progression—driven by genetic susceptibility, microenvironmental factors, or systemic influences that also promote recurrence and metastasis. Timing matters too: patients who develop a contralateral tumor within the early years after their primary diagnosis fare significantly worse than those with a longer disease-free interval, suggesting more aggressive biology or an inadequate response to initial systemic therapy, and arguing for closer monitoring and potentially more intensive adjuvant treatment in high-risk patients.

These findings collide directly with one of the most contested decisions in modern breast surgery: contralateral prophylactic mastectomy, the removal of the unaffected breast to prevent a second primary. The procedure has surged in popularity, particularly among young women with early-stage disease, driven by perceived survival benefits and psychosocial concerns. The evidence, however, is sobering. While prophylactic mastectomy effectively lowers contralateral incidence, a meaningful survival benefit appears confined to high-risk individuals—those with BRCA1/2 mutations or a strong family history. For most patients with sporadic breast cancer, the dominant threat to survival is distant metastasis from the original tumor, not a new cancer in the opposite breast, so the risk reduction does not translate into longer life. Meanwhile the surgery carries real costs: chronic pain, lymphedema, reduced range of motion, loss of sensation, and the burden of reconstruction, which may require multiple operations and prolonged recovery. Nonsurgical alternatives such as prophylactic contralateral breast irradiation remain of uncertain effectiveness. The review stresses that decisions should rest on evidence-based risk assessment and shared decision-making rather than fear.

The psychological dimension threads through every stage of this landscape. For many survivors, fear of a second cancer is a persistent source of anxiety that shapes treatment choices, relationships, self-perception, and daily functioning, especially among younger and high-risk patients. Surveillance itself imposes burdens: beyond annual mammography, the standard imaging modality, high-risk women may be offered supplemental screening with digital breast tomosynthesis, contrast-enhanced mammography, ultrasound, or magnetic resonance imaging, and liquid biopsy is being explored as a future tool for early detection. Young survivors may also warrant work-up for hereditary cancer predisposition syndromes, with positive findings triggering guideline-based screening at other sites. Each additional test brings potential false positives, costs, and emotional strain, which is why the review advocates personalized surveillance calibrated to individual risk rather than one-size-fits-all intensification.

The overarching message is one of balance. Contralateral breast cancer has become less common thanks to endocrine therapy, targeted agents, and better screening, yet when it occurs it carries profound prognostic implications, and the tools for predicting it—though improving—remain imperfect. Risk stratification models should be applied to every survivor as part of integrated follow-up care, preventive strategies should be weighed against their impact on quality of life to avoid overtreatment, and psychosocial support should be treated as a core component of management rather than an afterthought. As genomic prediction, artificial intelligence, and novel imaging mature, the goal, Hosseini argues, is a multidisciplinary, patient-centered framework in which the right survivors receive the right intensity of prevention and surveillance—and no one undergoes unnecessary surgery on the basis of fear alone.

Subject of Research: Risk assessment, prevention, and outcomes of contralateral breast cancer in breast cancer survivors

Article Title: Rethinking contralateral breast cancer: latest insights on risk, prevention, and outcomes

Article References: Rethinking contralateral breast cancer: latest insights on risk, prevention, and outcomes. (n.d.). https://doi.org/10.1007/s44272-025-00030-6

Image Credits: AI Generated

DOI: 10.1007/s44272-025-00030-6

Keywords: contralateral breast cancer, breast cancer survivors, BRCA1/2 mutations, endocrine therapy, prophylactic mastectomy, risk prediction models, hormone receptor status, cancer surveillance, triple-negative breast cancer, liquid biopsy, SEER database, second primary malignancy

The scale of the problem the study addresses is considerable. According to the Global Cancer Statistics 2020 report, roughly 184,615 people worldwide are diagnosed with laryngeal cancer each year, and nearly 100,000 die from it annually. In China alone, projections for 2022 estimated almost 30,800 new cases and about 16,900 deaths. Despite an expanding arsenal of treatments—surgery, radiotherapy, chemotherapy, and combinations tailored to individual circumstances—the five-year overall survival rate has stubbornly hovered between 50 and 60 percent for years. That stagnation, the authors argue, reflects in part the limits of the tools clinicians use to judge prognosis. The TNM staging system, which classifies tumors by their size and local extent, lymph node involvement, and distant metastasis, performs reasonably well when applied to entire populations but falters at the level of the individual patient, where two people with ostensibly identical stage designations can experience dramatically different outcomes.

To build a more discriminating instrument, the researchers turned to the Surveillance, Epidemiology, and End Results database, a vast population-based registry maintained by the United States National Cancer Institute. From records spanning 2000 to 2020, they identified 3,866 patients aged 18 or older with confirmed laryngeal squamous cell carcinoma, applying strict inclusion criteria based on International Classification of Diseases site and histology codes and excluding patients with multiple primary tumors, missing data, diagnoses made only at autopsy, or survival times shorter than three months. The cohort was then randomly split, by computer-generated assignment, into two equal halves: a training cohort of 1,933 patients used to identify prognostic factors and construct the model, and a validation cohort of another 1,933 patients used to test whether the model’s predictions held up in independent data.

The statistical machinery behind the nomogram began with univariable Cox regression analyses, which screened each candidate variable for an association with cancer-specific survival—defined as the interval from diagnosis to death specifically attributable to laryngeal cancer. Variables passing that initial filter were then subjected to multivariable Cox regression, a technique that isolates the independent contribution of each factor while accounting for the influence of all the others. Ten variables emerged as significantly associated with survival: age, race, marital status, primary tumor site within the larynx, tumor size, histologic grade, and the T, N, and M stages of the TNM system, along with whether the patient had undergone surgery on the primary tumor. Each of these factors was assigned a weighted point value in the final nomogram, allowing a clinician to total a patient’s score and read off the estimated probability of surviving one, three, or five years.

The demographic portrait of the cohort underscores the epidemiology of the disease. The overwhelming majority of patients were men over the age of 50—91.83 percent of the sample fell into that age bracket—consistent with the male-to-female incidence ratio of roughly seven to one reported for laryngeal cancer globally. Nearly 80 percent of patients were white, and tumors arose most frequently in the supraglottis, the region above the vocal cords, which accounted for 49.46 percent of cases, followed by the glottis, the vocal cord region itself, at 37.84 percent. Most tumors were moderately differentiated, classified as Grade II, and the vast majority of patients had no lymph node involvement or distant spread at diagnosis. Where distant metastases did occur, the lung was by far the most common site, hosting 64.5 percent of such lesions. During follow-up, 851 patients died of laryngeal cancer, split almost evenly between the two cohorts, with a median survival time of 32 months overall.

The performance metrics of the finished nomogram tell a story of solid, reproducible accuracy. Harrell’s concordance index, or C-index, which measures how well a model ranks patients by risk, reached 0.709 in the training cohort and 0.734 in the validation cohort—figures indicating substantial discriminatory capability. Receiver operating characteristic analysis, which evaluates the trade-off between sensitivity and specificity, yielded area under the curve values of 0.728, 0.746, and 0.766 for one-, three-, and five-year survival predictions in the training cohort, and 0.771, 0.751, and 0.763 in the validation cohort. Calibration curves, generated through 1,000 bootstrapped resampling iterations, showed close alignment between predicted and observed survival probabilities in both cohorts, meaning the model’s confidence in its estimates was matched by real-world outcomes.

Perhaps most striking is how the nomogram fared against the entrenched TNM staging system in head-to-head comparisons. Three complementary statistical approaches all pointed in the same direction. Decision curve analysis, which quantifies the net clinical benefit of a predictive model across a range of decision thresholds, showed the nomogram delivering greater net benefit than TNM staging across a broad spectrum of threshold probabilities in both cohorts. Net reclassification improvement, which measures how many patients are correctly shifted into better or worse risk categories compared with a reference model, reached 0.205, 0.304, and 0.322 at one, three, and five years in the training cohort, and 0.163, 0.203, and 0.182 in the validation cohort. Integrated discrimination improvement, a related metric capturing gains in predictive precision, was likewise positive at every time point. Together, these results suggest the nomogram genuinely adds information that TNM staging alone cannot provide.

Beyond its predictive scores, the tool offers a practical route to risk stratification. Using X-tile software, the researchers established cutoff values of 146 and 247 total points, dividing patients into low-, intermediate-, and high-risk groups. Kaplan-Meier survival curves, analyzed with the log-rank test, revealed starkly divergent survival trajectories among the three strata in both cohorts, with high-risk patients experiencing significantly worse cancer-specific survival. In effect, the nomogram converts a scatter of individual clinical details into a single number that sorts patients into clinically meaningful risk tiers—a capability that could guide decisions about treatment intensity and follow-up frequency for individual patients rather than average ones.

The study’s findings on individual prognostic factors echo and extend earlier research. Older age carried a steeply rising mortality risk, with hazard ratios of 1.82, 2.01, and 3.57 for patients aged 50 to 59, 60 to 69, and 70 and above, respectively, compared with younger patients—a pattern the authors attribute to the cumulative accumulation of environmental exposures over a lifetime and to age-related changes in laryngeal tissue. Black patients faced a significantly elevated risk of death, with a hazard ratio of 1.47, a disparity the researchers link to a complex interplay of genetic predisposition, socioeconomic factors, unequal access to screening and treatment, and differences in health behaviors and environmental exposures. Notably, male sex, often reported as a risk factor in prior studies, did not reach statistical significance here, with a hazard ratio of 1.22 and a P-value of 0.1050—a discrepancy the authors suggest may stem from the limited number of female patients in the cohort.

The authors are candid about the limitations that temper their conclusions. The model was built exclusively on retrospective data from the American population, and the researchers caution that it is not yet mature enough for direct application in Chinese patients given ethnic differences. Crucially, some of the most important risk factors for laryngeal cancer—smoking habits, alcohol consumption, and human papillomavirus infection—could not be incorporated because the SEER database does not capture them, leaving open the possibility that the model’s accuracy could improve further with richer inputs. Prospective data collection, larger and more geographically diverse cohorts, and external validation across different populations remain essential next steps. Still, the study represents a meaningful advance: a convenient, practical, and well-calibrated tool that gives clinicians a sharper lens on individual prognosis at a time when the survival outlook for laryngeal cancer patients has barely moved in decades. For a disease that strikes hardest at older men and disproportionately burdens disadvantaged communities, a better way to see who is truly at risk may be the first step toward changing the numbers themselves.

Subject of Research: Development of a nomogram for predicting cancer-specific survival in laryngeal squamous cell carcinoma using SEER database data

Article Title: A nomogram for predicting cancer-specific survival in patients with laryngeal squamous cell carcinoma: a population-based study from SEER database

Article References: Zhou, H.-L., & Chen, D.-D. (2025). A nomogram for predicting cancer-specific survival in patients with laryngeal squamous cell carcinoma: a population-based study from SEER database. Clinical Cancer Bulletin, 4(1), Article 8. https://doi.org/10.1007/s44272-025-00037-z

Image Credits: AI Generated

DOI: 10.1007/s44272-025-00037-z

Keywords: laryngeal squamous cell carcinoma, nomogram, cancer-specific survival, SEER database, TNM staging, prognostic factors, Cox regression, risk stratification, head and neck cancer, decision curve analysis, C-index, survival prediction

The research team, led by Peiling Dai of the First Affiliated Hospital of Wenzhou Medical University along with colleagues across several Chinese institutions, focused on a group the authors abbreviate as EFLCBM: elderly female lung cancer patients with bone metastases. Rather than relying on broad staging categories that lump together wildly different patients, the team set out to build a nomogram—a visual scoring instrument that combines multiple clinical variables into a single personalized probability of survival. Nomograms have become increasingly popular in oncology precisely because they translate complex statistical models into something a physician can use at the bedside, weighing tumor characteristics, demographics, and treatment factors to estimate an individual patient’s risk.

The foundation of the study is the Surveillance, Epidemiology, and End Results database, the large population-based cancer registry maintained by the United States National Cancer Institute. From SEER, the researchers identified 3,194 elderly women with lung cancer that had metastasized to bone. To guard against the well-known pitfall of models that perform well only on the data used to build them, the team split the SEER cohort into a training set and an internal validation set, then added a genuinely independent check: 71 patients treated at their own medical institution formed an external validation cohort. This three-tiered structure—train, validate internally, validate externally—is considered the gold standard for demonstrating that a prediction tool has real-world rather than merely mathematical staying power.

Statistically, the work rested on Cox proportional hazards regression. The researchers first ran univariate analyses to screen candidate variables associated with overall survival, then multivariable analyses to isolate the independent prognostic factors—those variables that retained predictive power even after accounting for everything else. The significant factors were then assembled into the nomogram, with each variable assigned a weighted score reflecting its contribution to survival risk. The endpoint the model targets is overall survival, estimated at 12, 18, and 24 months after diagnosis, giving clinicians a rolling window of personalized prognosis rather than a single static number.

The performance metrics are where the study earns its credibility. The area under the curve, or AUC, a standard measure of a model’s ability to distinguish patients who survive from those who do not, reached 0.752, 0.766, and 0.786 for 12-, 18-, and 24-month survival respectively in the training cohort. Values above 0.75 are generally regarded as indicating useful discrimination, and the model held up well beyond its own training data: AUCs of 0.743, 0.727, and 0.716 in the internal validation cohort, and 0.679, 0.633, and 0.690 in the external validation cohort. The drop in the external cohort is expected and honest—no prediction model transfers perfectly between institutions—but the figures suggest the tool retains meaningful signal when applied to patients it has never seen.

Discrimination alone, however, is not enough. A model can rank patients correctly yet still produce biased absolute probabilities, so the team examined calibration curves, which plot predicted survival against observed survival. Across the SEER cohorts and the external institutional cohort, the calibration curves showed that the model’s predictions tracked closely with what actually happened to patients, a sign that the risk estimates are not merely rank-ordered but quantitatively trustworthy. The researchers also applied decision curve analysis, a relatively recent technique that quantifies the clinical net benefit of acting on a model’s predictions across a range of threshold probabilities. Here the nomogram delivered perhaps its most persuasive result: across a wide span of clinically relevant thresholds, acting on the model yielded more net benefit than treating everyone, treating no one, or relying on simpler default strategies.

Beyond prediction, the study converted the nomogram into a practical risk classification system, stratifying patients into groups whose survival trajectories differed markedly on Kaplan-Meier analysis. This stratification then served a second, therapeutically significant purpose: identifying which treatment modality best serves which risk group. The subgroup analysis produced a clear pattern. For patients classified as low risk, chemotherapy emerged as the recommended mainstay of treatment, while for high-risk patients, the early use of radiotherapy appeared more beneficial. In a field where treatment of elderly patients with metastatic disease has often been guided by one-size-fits-all instincts—or by therapeutic nihilism—this finding offers an evidence-based framework for tailoring intensity and modality to the individual.

The implications extend beyond the specific population studied. Elderly women with lung cancer have historically been underrepresented in clinical trials, leaving their clinicians to extrapolate from evidence gathered in younger, fitter, and predominantly male cohorts. A tool built explicitly from and for this population closes part of that gap. It also arrives at a moment when radiotherapy and systemic therapy for bone metastases are both evolving rapidly, making the question of sequencing—who gets radiation first, who gets drugs first—more pressing than ever. The authors’ risk-based framework gives oncologists a principled starting point for those conversations, potentially sparing low-risk patients the side effects of treatments unlikely to help them while accelerating high-risk patients toward interventions that may palliate pain and prevent skeletal complications.

The study, which was retrospectively conducted and therefore exempted from individual informed consent by the ethics committee of the First Affiliated Hospital of Wenzhou Medical University, does carry the inherent limits of registry data, and the modest size of the external cohort of 71 patients means the external validation should be replicated elsewhere. Still, the authors conclude that their nomogram demonstrates good discrimination, calibration, and clinical benefit, and that the accompanying risk classification system successfully identifies the best beneficiary populations for radiotherapy and chemotherapy among elderly female lung cancer patients with bone metastases. For a growing group of patients whose prognosis has long been described in grim generalities, the message of this research is refreshingly specific: with the right data, an individualized forecast—and an individualized treatment plan—is now within reach.

Subject of Research: A prognostic nomogram for predicting survival outcomes in elderly female lung cancer patients with bone metastases

Article Title: Development and validation of a nomogram for predicting disease outcomes in elderly females with lung cancer metastatic to the bone

Article References: Dai, P., Chen, S., Yang, B., Chen, K., Huang, Z., & Liu, J. (2026). Development and validation of a nomogram for predicting disease outcomes in elderly females with lung cancer metastatic to the bone. BMC Cancer. https://doi.org/10.1186/s12885-026-16991-3

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16991-3

Keywords: lung cancer, bone metastases, nomogram, prognosis, SEER database, chemotherapy, radiotherapy, elderly women, overall survival, decision curve analysis, risk classification, BMC Cancer