David Spetzler
Analyst · Guggenheim
Thanks, Brian. I will walk through the 2 new products and then the foundation model behind our digital AI platform, our planned expansion of our ACHIEVE program for Caris Detect and will close on interception. Everything I'm about to talk through comes back to what Dave said at the open, deeper, not cheaper. Depth and breadth are what allows us to complete a comprehensive approach across the patient journey from early-stage detection to therapy selection and now monitoring and MRD. Slide 11 frames the 2 launches of the fourth quarter together. Caris Assure Therapy Response Monitoring, which launched recently, covers the patient on therapy. Caris Confirm, launching in December, covers surveillance after treatment. Both are built on platforms we already run. Monitoring sits on our whole-exome platform. Caris Confirm sits on our whole-genome platform with the initial tissue sample to define the signature, then follow-up testing in blood. Each has its own evidence base, MONSTAR and our real-world cohort for monitoring, ENSEMBLE and the real-world cohort for Caris Confirm. Reimbursement work is underway on both, a MolDX submission and preparation for monitoring and a cancer-type-by-cancer-type strategy for Caris Confirm, and I will discuss each a little more. Let's start with Caris Assure Therapy Response Monitoring, which launched recently as part of the upgraded Caris Assure platform. For therapy selection, Caris Assure tells the oncologists which drug to choose, for monitoring the same whole-exome liquid workflow with matched buffy coat to remove clonal hematopoiesis gives a quantitative tumor fraction and serial draws classify the patient as a molecular response, stability or progression. The readout arrives roughly 4 to 8 weeks ahead of the scheduled scan. The validation is on the slide. Clinical validation included 124 serial pairs and in the 90-patient landmark cohort, 96% of patients classified as molecular responders went on to show disease control on imaging, 49 out of 51. Patients classified as molecular progression had a hazard of subsequent radiographic progression nearly 9x higher than responders, 8.99 or 7.16 after statistical adjustments. And in an early independent real-world cohort of 1,037 patients, molecular progression carries a 2.35x hazard versus response with the 3 categories separating in order. Analytically, the assay detects tumor fractions down to 0.1% with precision under 9% CV across LoDs and runs. Moving to Slide 13, Caris Confirm is our entry into minimal residual disease and is currently planned for launch in December. The approach is tumor informed and whole genome. We sequence the tumor and a matched normal across the entire genome to define a patient's set of trackers. A tracker search base of roughly 2.5 billion bases. We then track those variants in serial blood draws on the same whole-genome platform that powers Caris Detect. The pathway is relatively simple. A baseline blood draw with buffy coat plus tumor tissue from surgery, tumor-normal subtraction to define the patient's tracker set, a landmark blood draw 3 to 6 weeks after surgery, and then surveillance draws every 3 to 4 months. That is what gives us an ultra low limit of detection by design. The analytic performance speaks for itself with a minimum positivity threshold of 1.38 parts per million. The intended use is pan-tumor, Stage 1 through 3 solid tumors, after curative intent treatment, a setting where patients are monitored for years. So like therapy monitoring, this is a reoccurring test on infrastructure that is already in place. Moving to Slide 14, this is our engine under our digital AI platform. HERO is our pathology foundation model. It was trained on 248,606 routine H&E slides across 52 cancer lineages and 98% of those slides, more than 243,000, also have RNA sequencing alongside DNA sequencing and IHCs. That pairing is the asset. The molecular results give us every slide as labels. The process runs left to right on the slide, the H&E is scanned at 20x. The tissue is cut into tiles about 1/10 of a millimeter across, roughly 2,300 per slide. HERO describes each tile as 3,072 separate numbers, the same for every target with no retraining. For each target, the model learns how much each tile matters and weights the tiles into 1 summary of the slide. And from that, 3 models, 1 for each of DNA, RNA and IHC, predict 297 alterations, 368 RNA expression levels and 21 protein stains from the same slide alone. Our model powers Caris digital AI applications. That is why the pipeline on the right, morphological subtype, tumor grading, mutation prediction, treatment response and survival prediction can move quickly. Each is a new head on the same foundation trained on data we already have. Now let me turn to Caris Detect. As we have discussed before, Slide 16 shows why we built Caris Detect to combine many signals across cell-free DNA and RNA, expanding our shots across goals, expanding our shots on goal across each early tumor. Moving to Slide 17, ACHIEVE 1, the basis for our commercial launch, showed 99.2% specificity and 60.3% sensitivity in Stage I and Stage II cancers, the stages where the cure rate is highest, and I will give an update on our planned next readout from ACHIEVE 2 in a few moments. Now as part of this quarterly update, I will provide an update on our ACHIEVE program. As you know, this is the clinical evidence program behind Caris Detect and one that we are announcing an expansion to, with 4 studies under the ACHIEVE name totaling 63,000 patients. ACHIEVE 1 is the 3,000-patient study you just saw. That readout was delivered in the first quarter. ACHIEVE 2 is our large-scale validation with a target of 25,000 participants. We have enrolled approximately 20,000 patients, 80% of the way there, and we certainly expect an interim readout in Q1 of 2027 as we have begun to run these samples. ACHIEVE 3 is a 10,000-patient prospective study in a high-risk population with results returned to patients and a repeat Caris Detect at 12 months. And ACHIEVE 4 is a 25,000-patient prospective study in a general risk population, also with results returned. ACHIEVE 3 and 4 are in planning now. And on Slide 19, as DDH discussed, we have launched 2 campaigns for Caris Detect, and both point consumers to Caris Detect, and these will continue to be rolled out across markets into Q4 and beyond. Finally, personalized mutation interception, we outlined this program on our last call. So today, I will keep it brief and focus on how the pieces of our platform fit together to support it. As a reminder, our personalized vaccine approach is built on capabilities we already run. Caris Detect finds the early signal, our sequencing defines the mutations that inform a personalized vaccine and serial blood draws track whether it is clearing. Slide 21 shows the step where our platform is most differentiating, choosing the right mutation. We select only the mutations that are expressed and visible to the immune system as vaccine targets then track response over time. This is where our dataset does the work. Our ranking model achieves 83.8% positive predictive value and 86.5% for the top-ranked variant per patient, measured against a paired biopsy cohort of more than 42,000 patients. So we focus on the mutations most likely to prime a response. So that is the focus of our science. Two new reoccurring use products on the platform, a foundation model that lets digital AI portfolio grow quickly, detection that is both broad and precise with an expanded 63,000-patient evidence program behind it and an interception strategy that is novel, all resting on the depth-first approach. With that, I'll turn it over to Luke.