AI Scribes and Agents for Oncology Practices

• The ambient scribe. First. Oncology notes are among the longest in medicine and the after hours burden is heavy. Measure documentation time per new consult and per treatment visit for four weeks before go live.
• Drug prior authorization. A close second and in some practices first. Every high cost oral and infused agent, most imaging and many supportive drugs require authorization, often with strict criteria tied to line of therapy, biomarker status and prior regimens. The note the scribe produces is the source document; the authorization tool assembles and tracks the request. The CMS prior authorization rule is shortening payer timelines for regulated plans, which moves the bottleneck to the practice. Price yours with the prior authorization cost calculator.
• Care coordination. Oncology patients see surgeons, radiation oncologists, palliative care, infusion, pharmacy and primary care. Tools that draft the referral and update letters from the visit and track pending steps reduce the coordination load on nurse navigators.
• Symptom message triage. Patients on treatment message about symptoms that may be urgent. Routing to the triage nurse promptly is valuable; any tool that judges urgency itself is not acceptable here.
• Infusion and visit scheduling. Chair time, lab timing, physician visit and pharmacy preparation have to align. Scheduling optimisation is a real project with a real return, and a different product category from a reminder tool.
• Coding. Long visits, prolonged services, drug administration codes and complex E/M. Review tools help; auto coding does not belong here.

Oncology is the specialty with the most specific vendor investment in the registry, checked September 2026.

DeepScribe positions oncology as its strongest area: a dedicated oncology specialty page, a published adoption by New York Cancer and Blood Specialists, and a stated volume of cancer care visits feeding its models. Its EHR integrations are API based on Epic, athenahealth, eClinicalWorks and NextGen; ask specifically about oncology EHRs and treatment plan modules. Pricing not published; see DeepScribe pricing.

Abridge is live at UPMC across more than forty specialties including oncology, per its press releases, and is native on Epic. Ambience names oncology among the complex domains it supports in funding announcements and third party coverage, though its homepage lists settings rather than specialties. Suki lists Oncology and Hematology in its supported specialties table. Heidi features a medical oncology testimonial on its specialties page and has a free tier. Pricing for all four is not published except Heidi's free tier; see the Abridge, Ambience and Suki pricing pages.

Dragon Copilot claims broad coverage and is deployed in cancer centres through health system agreements without naming oncology on a product page. Freed, Nabla and Sunoh.ai do not name oncology. Oncology specific EHR vendors also ship their own documentation AI; those products sit outside our registry and belong on a cancer centre's shortlist. Our independent scribe comparison explains the weighting.

Treatment plan precision. Regimen name, each agent, dose and units, schedule, cycle number, intent (curative, adjuvant, palliative), and any dose modification with its reason. Test the scribe against your own regimens in the pilot and read every plan section, because the failure modes, unit errors, agent confusion between similar names, dropped modifications, are safety events. The scribe documents the discussion; the order lives in the treatment plan module with its checks, and nothing should flow automatically from one to the other without pharmacist review.

Toxicity and response. Grading of adverse events and response assessment should be captured in the terms the practice uses, usually CTCAE grades and RECIST or equivalent. Confirm the scribe writes the grade the oncologist states rather than a paraphrase.

Clinical trials. Trial participants' visits generate source documentation that a sponsor and an IRB will audit. Recording the visit and generating a note is usually acceptable if the protocol and consent permit it, but the transcript is now a source document too, subject to retention and inspection requirements that differ from routine care. Involve the research office before recording any trial visit, and ask the vendor how retention can be set per patient. HHS guidance on research uses of health information and the Common Rule apply alongside HIPAA.

Tumour boards. A multidisciplinary conference discussing dozens of patients is not an encounter, and recording it with an ambient scribe raises consent and confidentiality problems for every patient discussed. Tumour board summaries should come from the conference's own documentation workflow. Some vendors offer meeting summarisation as a separate product; evaluate it as a separate decision with its own governance.

Goals of care and advance care planning. These conversations have their own billing codes based on time and their own documentation elements. The scribe can capture them well, and the oncologist should confirm that time based elements are recorded accurately rather than inferred.

The standard verbal consent at the start of the visit, documented, with written notice at check in, built to all party consent rules. Oncology adds weight to each element.

Family members are almost always present and often speak. Everyone speaking should consent, and the note should attribute history and questions appropriately. Prognosis and end of life conversations are the ones patients most often do not want recorded; the oncologist needs a pause that works and the habit of offering it. Refusal must be frictionless and must not change the care.

Genetic and genomic information appears in most consults and carries additional protection under GINA and state genetic privacy laws. A transcript containing germline results held by a vendor is a sensitive artefact; set retention to the minimum, decline model training in the contract, and confirm subprocessor locations. Our HIPAA and AI page covers the business associate agreement and the clauses that matter, and the scribe compliance checklist the operational steps.

Where the practice sends AI drafted messages to patients, state disclosure laws such as California's apply. Any tool that touches utilisation decisions, including drafting authorization appeals that argue medical necessity, should be reviewed by a clinician before it leaves the practice; some state laws now constrain AI in utilisation review on the payer side, and the provider side should hold itself to the same standard.

Use the AI scribe ROI calculator with your own figures; these are illustrative.

• Ten medical oncologists, 18 visits a day, 220 clinic days a year, of which about four visits a day are new consults or treatment planning.
• Measured documentation of 20 minutes per consult and 8 minutes per treatment visit, giving about 192 minutes a day.
• A 40 percent reduction, returning 77 minutes a day.
• Loaded oncologist cost of 175 USD per hour.
• Licence cost of 450 USD per oncologist per month, an assumption in the absence of published specialty pricing.

Returned time is about 282 hours per oncologist per year, worth roughly 49,000 USD, or 494,000 USD across ten against 54,000 USD of licences. Even at half the reduction the case clears easily. The authorization number is separate and large: a practice submitting 600 drug and imaging authorizations a month at 30 staff minutes each spends 3,600 staff hours a year on submission, before appeals and before the treatment start dates lost to delay. The prior authorization cost calculator prices that.

Oncologist burnout and turnover are well documented and replacement takes a year or more in most markets; the clinician turnover cost tool with your own replacement figure will show that one retained oncologist covers the entire programme.

• Show us a redacted new consult note for a common tumour type. Are regimen, agents, doses, schedule, cycle and intent recorded exactly, and are dose modifications and toxicity grades captured in our terms?
• Confirm nothing flows from the note into the treatment plan or orders without pharmacist review.
• How is retention set per patient, so trial participants can be handled under their protocol?
• Does the same recording draft the referring physician letter, the patient summary and the authorization narrative?
• How does the pause work for prognosis conversations, and what does the note show?
• Which oncology practices on our EHR and treatment plan module will take a reference call?
• Retention, training, subprocessors and genetic information handling in contract language. Our vendor questions guide covers the rest.
• Pricing at ten oncologists and at thirty, and the exit term.

Oncology is the specialty where the scribe case is easiest to make and the deployment is hardest to get right, because the note feeds the pharmacy, the payer and the sponsor, and each has a different standard. Vendor demos will show a beautiful consult note. The pilot has to show regimen precision across your own regimens, the trial retention question answered and the authorization team reading the notes.

Our vendor selection service evaluates scribe and authorization vendors together, runs a pilot with your pharmacist and authorization lead scoring the notes, and negotiates retention, training and genetic data terms. Our governance and compliance service builds the trial and tumour board policy that your research office and cancer committee will need. We take no vendor commissions. Book a call with your visit mix, EHR, treatment plan module and monthly authorization volume.