05/14/2026
Dennis Horvath gave my patient a radiation burn. Then he hid the evidence under lead.
Margaret Hsu is sixty-seven years old. She came to Riverside Cancer Pavilion for liver treatment — stereotactic body radiation therapy, five fractions, precise geometry, tight margins. She left with moist desquamation on her torso. Radiation-induced tissue injury. The kind the burn clinic photographs and calls grade three. The kind that heals slowly, in stages, with wound care appointments and a compression sleeve she wears at church with a floral blue pattern.
The machine that treated Margaret drifted. TrueBeam Unit 3 overshot its prescribed output by 2.4% on August 12. 2.7% on August 13. 3.1% on August 14. The institutional action threshold is 2.0%. A machine hitting that number is supposed to stop. It is supposed to alert the physics team, pause the treatment queue, and trigger a calibration review before the next patient touches the couch.
The machine did not stop. The physicist deleted the logs.
My name is Rochelle Vause-Adekunle. I am the Radiation Safety Officer at Riverside Cancer Pavilion. I have been badging into these vaults for twelve years. I am a Certified Health Physicist candidate. I hold the ARRT advanced imaging credential. I run monthly output constancy checks — ion chamber, calibrated electrometer, bias voltage verified before every session, readings logged to three decimal places. I maintain the radiation safety committee minutes. I file incident reports with the Virginia Department of Health under the radioactive materials license conditions. When something goes wrong — I file. When something goes wrong and someone tries to hide it — I still file.
Dennis Horvath was the Lead Medical Physicist. He was my mentor. He taught me isodose curves on a yellow legal pad — freehand, smooth ellipses that matched the calculated distribution within a pencil width. He kept a Pentel 0.5mm pencil behind his ear, and the clip left a dent in his cartilage. I have a matching dent from twelve years of the same habit. He said "never outrun physics" and meant it. He ran half-marathons for the children's oncology charity. His race bib — "DoseRightDennis," number 4471, fabric frayed at the hem from concrete wear — was pinned to the break room corkboard the same week Margaret Hsu was on the treatment couch. He bought me decaf at morning meetings because he thought caffeine gave me tremor during ion chamber readings. He called me Rock for twelve years. The nickname used to sound like trust.
At three forty-five every night, the TrueBeam linac runs a self-test. The gantry rotates to zero. The jaws open to reference field size. The dose monitor calibrates to baseline. The machine logs everything — output readings, jaw positions, dose monitor values — and uploads the full event trace to the Varian manufacturer cloud automatically at 03:47, before the morning therapists arrive. I do a late walkthrough once a month. I check vault door interlock magnets, emergency stop button function, radiation safety signage. The Unit 3 display clock read 03:45 on my last walkthrough, two weeks before I found the gap. The LED ring around the gantry head flickered green, reflected on the epoxy vault floor in a circle of light. Dennis was in the control room. He saw me on the vault camera feed. "She knows the ghosts' shift," he said through the intercom. He meant the machine's automated routines — the tests that run when no one is supposed to be watching. 03:45 was folklore then. It was not folklore anymore.
On September 9, at six-twenty in the morning, I sat at the secondary physics workstation. I opened the MOSAIQ oncology information system — attachment index for TrueBeam Unit 3, sorted by service event date. August 12: daily output constancy file. August 13: daily output constancy file. August 14: blank row. August 15: blank row. No attachment. No event trace. One pixel of scrollbar gap where two 4.2-megabyte files should have been.
On the right screen: the Varian remote service cloud extract — obtained through a third-party record request to Siemens Healthineers, processed under the hospital's vendor data access clause. The cloud showed the August 14 trace, uploaded automatically at 03:47 by the TrueBeam Unit 3 service module during the nightly self-test. Complete. Checksummed. 4.2 megabytes.
The local copy was gone. The manufacturer's copy was not.
I checked the hash values. The file was not corrupted. It was not missing from a system glitch. The MOSAIQ downtime maintenance window is scheduled for Sunday at two a.m. The deletion event on the vendor's log corresponded to a Monday morning session — work hours, manual action, someone with local system access. Remote remembers what local forgot.
I also checked the Varian cloud for August 12 and 13. The cloud showed the true output readings for both days. 2.4%. 2.7%. Both above the 2.0% action threshold. Both days showed values within tolerance in MOSAIQ — 0.4% and 0.6%. The local records were truncated before they reached the attachment store. Three consecutive days of falsified QA data. Five SBRT liver cases cleared during that window. One of them was Margaret Hsu.
I requested Dennis Horvath's hiring file from HR — the original interview panel notes, lawfully accessible to the RSO under the hospital's credential review policy for safety-critical roles. Page six. In the margin, in Dennis's handwriting, initialed DH: "Never override vendor safety interlocks — policy and personal commitment." His own words from the day he was hired. His own words from before he overrode them.
An independent physicist, Dr. Yasmin Ali-Borges, reviewed the vendor trace data by encrypted screen share and recalculated the Unit 3 output for Margaret's SBRT fractions. The dose-volume histogram showed a bulge at the target boundary — a 3.1% delivery overshoot on August 14. "That's not variance," she said. "That's a burn." Her memo was four pages, peer-reviewed methodology, signed and dated. I encrypted it and stored it with the vendor extract and the HR file.
I walked to the hot lab sink. I printed the Varian trace PDF — all three days, timestamped, checksummed. I placed the printout on the lead brick beside the sink. I turned on the water. Too hot. I did not adjust the faucet. The soap dispenser clicked twice by mistake — I pressed it harder than I needed to. Thirty seconds. I dried my hands on the paper towel.
The printout sat on the lead brick. 03:47. August 14. Unit 3. 3.1% above baseline. Margaret's treatment date. Margaret's fraction. Margaret's skin.
I did not page Dennis.
The CMO drop slot was on the other side of the building. The FDA MedWatch form was on my secondary workstation. The VDH complaint portal was bookmarked.
None of that had happened yet.
PART 2 👇
