Cancer Splits Into Subtypes That Change the Prognosis

Two immune archetypes in head and neck cancer

Researchers analyzing 2,121 de-identified head and neck squamous cell carcinoma cases from the Tempus database identified two distinct immune archetypes, labeled Inflamed and Immune-Desert, with the Inflamed group highly enriched in high immune-phenotype-score tumors and in HPV16 positivity and combined positive score of at least 1. m:…4636

The survival gap between the archetypes is substantial: patients in the Inflamed archetype had median real-world overall survival of 25.1 months, compared with 16.8 months for the Immune-Desert archetype, and multivariate analysis confirmed high immune-phenotype-score status, HPV16 positivity, and combined positive score of at least 1 as independent prognostic factors for improved survival. m:…4636

The study's implication is pointed at the higher-risk group: the authors highlight the need for novel T-cell recruitment strategies for the Immune-Desert population, which frames the archetype not just as a label but as a guide to what kind of therapy research the group needs next. m:…4636

One TP53 mutation, a distinct pan-cancer subgroup

A second study analyzed 200,391 de-identified patient samples with solid tumors using the Tempus Lens Platform and found TP53-Y220C in 0.9 percent of tumors, totaling 1,778 cases, with ovarian cancer showing the highest prevalence at 3 percent, and concluded that the mutation defines a genomically and clinically distinct pan-cancer subgroup. m:…6946

Patients with the mutation were older than those with wild-type TP53 and less frequently stage IV, yet tumors with Y220C had shorter real-world overall survival than wild-type tumors, with a hazard ratio of 1.26, and the disadvantage grew when Y220C co-occurred with CDKN2A alterations, which carried a hazard ratio of 1.41 compared with Y220C alone. m:…6946

The co-alteration pattern is detailed: across solid tumors the most frequent co-alterations with Y220C were KRAS at 21 percent and CDKN2A at 23 percent, nearly all Y220C pancreatic tumors co-harbored KRAS, 80 percent of colorectal cancers with Y220C had APC alterations, and 59 percent of ovarian Y220C cases were homologous-recombination-deficiency positive, with significantly worse survival seen in ovarian, uterine, and pancreatic cancers but no survival difference in high-grade serous ovarian cancer. m:…6946

An AI panel for lung cancer screening

The third study turns from stratification to screening: Tempus developed a machine-learning-powered pre-screening panel to predict 12 clinically relevant biomarkers using standard stained whole-slide images from advanced non-small cell lung cancer, trained on multi-modal DNA sequencing and immunohistochemistry staining data from a cohort of 60,459 cases across 56,969 patients. m:…6053

The study used a 70 percent training, 15 percent tuning, and 15 percent holdout split, selected prediction thresholds targeting 90 percent sensitivity for each biomarker on the tuning data, and reported holdout sensitivity greater than 80 percent for all target biomarkers, including 92.49 percent for ALK fusion, 89.63 percent for KRAS, and 89.08 percent for PD-L1 at the 1 percent threshold. m:…6053

The authors present the panel as a way to reduce burdens of time, tissue, and cost in order to improve screening rates, which connects the three studies: whether the tool sorts patients by immune state, by mutation, or by predicted biomarker, the value lies in getting the right patients into the right next step sooner. m:…6053

Stratification as the throughline

The three cancer studies share one move: split a broad diagnosis into groups with different outlooks, whether by immune archetype, by a single mutation, or by predicted biomarkers, and then use the split to guide what happens next, from T-cell recruitment research for the high-risk immune group to faster screening for lung cancer patients. m:…4636 m:…6946 m:…6053

Scale is what makes the splits credible: cohorts of 2,121 head and neck cases, 200,391 solid-tumor samples, and 60,459 lung cases give the subgroup findings weight, while the detailed co-alteration patterns and holdout sensitivities show the work is measured rather than asserted. m:…4636 m:…6946 m:…6053