The Compassionate Channel

Confession first: I believed the brochure. For the first four years I believed it, and I want that on the record, because the record is the only thing I have left, and a record should have a true beginning.

The brochure is still in the archive if you ever want to see it. Corporate blue, a dove that was probably meant to look hopeful and reads as overcast. The sentence that mattered: “Our expanded access program provides investigational therapy to patients with no remaining options, at no cost.” Expanded access is the clinical name. The old word is compassionate use. Both mean the same thing: a dying person may receive a drug that has not been proven, outside the trial, for free, with no promises. The drug has not been proven. That is what “investigational” means. What the brochure left out was the arrow pointing the other direction, but the arrow was in the regulation all along. The same paragraph that built the channel said the channel must not interfere with the trial. They put mercy and evidence in the same sentence and hired someone to stand between them. That person was me, and I stood between them for twenty years.

I was hired to read consent forms out loud and answer a phone. The desk was a cubicle at the end of a corridor with a two drawer cabinet, and the cabinet held the log, because I started in the paper era and I was stubborn about it. Twenty years later the log was the only paper document in the building.

You need the machinery, so here is the machinery. The Sponsor had a drug. Cells carry a switch called KRAS, and in a common family of lung cancer the switch is jammed in the on position, so the cell divides like it has no boss. The drug was built to fit that one mutant copy of the switch and hold it closed, wedged in the off position. It worked. The problem was never the switch. The problem was proof.

To sell the drug, the Sponsor had to run a trial, and the trial had two arms, like a scale. One arm got the drug. The other got the current best treatment, which for that disease was a taxane and a prayer. Patients were assigned to arms by chance, and the whole religion of the enterprise is that chance does the housekeeping: given enough people, the two arms come out equal in everything that matters, and then the only difference left is the drug, and then you can know. The bar in the trial charter was a two sided p value under 0.005, which means, roughly: the odds that you would see a difference this big if the drug did nothing and the two arms were secretly the same crowd. Every word in that sentence is doing work, but this is the word that matters: conditional. The p value is conditional on the randomization having held. The randomization held.

The trial ran alongside the channel. Same patients, same disease, same desperation, minus the coin flip. The trial offered a fifty fifty chance at the drug. The channel offered the drug itself, if you got past me.

The first application I ever took was Delphine Abara. Her daughter drove her. She held the consent form in both hands, like paper might get away from her. I read her the side effects out loud, because the job was mostly reading things out loud. The liver: the drug can kill your liver, so we drew blood weekly. The gut: diarrhea, which is a brutal word to say to a woman who was trying so hard to hold on to something solid. And fatigue. She said, “Is that all?” and I said, “That’s all,” and she signed.

EP 0001. ABARA, DELPHINE. 61. NSCLC, STAGE IV, KRAS G12C. PS 1. APPROVED. REASON: “NO REMAINING OPTIONS. PLEASE.”

Six months later she met me at the desk with her labs in her own hand and she was steadier. Steadier is not a miracle; it is a drug working. I felt like the pivot point of the universe. I don’t have a better word for it than that.

The log had a reason column, because the Sponsor wanted to know why the desk did what it did. For years I wrote real things in it. “Chemotherapy exhausted. Family request. Please.” The record of what I believed is in that column, in my first hand. Year four, the column changed. I started writing “Meets criteria,” which is a sentence that says nothing, because the criteria were a weather system, and weather systems do not have reasons.

On paper the criteria were legible and published: performance status, which is how well you can walk and work on a scale of zero to four. Organ function. Prior lines of treatment. In practice they were a weather system, and the weather forecast came from the committee.

The committee met annually, four of us. The medical director spoke mostly about patients and meant mostly the shareholders. The clinician spoke mostly about the standard of care and meant the trial. Dr. Brandt said very little and wrote in a small black book, and for years I believed she was counting the cost of the people we admitted. I had the direction of her bookkeeping wrong, and I did not find out how wrong until the week after the log was scanned. I am telling this in the order I learned it, which is not the order it happened. I am not going to fix that.

Andre was not on the committee. Andre attended sometimes, said almost nothing, and when he spoke the room went quiet.

Once a quarter there was the accrual call, which was the most important document processing ceremony in oncology. When the trial was filling on schedule, the mercy door stood open and I could admit. When the trial was starving, the door closed and I became a rationer. The logic was always the same sentence: expanded access must not compromise the evidence that will bring this therapy to everyone. Say it with an open door and it means: be merciful, we have room. Say it with a closed door and it means: keep the control arm full. There is literature about compassionate use stealing enrollment from trials. The Sponsor’s lawyers had read that literature, and the Sponsor’s lawyers had built the schedule.

Nobody ever rolled up their sleeves and said we were rationing the dying so the control arm could fill, because they did not have to. The sentence did it for them. Dr. Brandt said it once, at a committee, the only strong thing I ever heard her say: “Mercy is not the opposite of this machine. Mercy is the most efficient component.” The medical director smiled like she had made a joke. She had not made a joke.

The admission I keep: Imani Diallo, year eleven. Nine years before the drug would read out. Imani was a nurse and it showed. Her records came in a binder with tabs. Her doctor called the Sponsor’s medical director, which is the difference between the sidewalk and the top floor. PS 1, labs like a textbook, and the quarter was green, green enough that even the people without tabs were getting past us that month. She asked me, in the intake room, with the north window, “Is it true about the liver?” I said, “We draw blood every week. I’ll be the one who calls you when the numbers even twitch.”

EP 2711. DIALLO, IMANI. 54. NSCLC, STAGE IV, KRAS G12C. PS 1. APPROVED. REASON: MEETS CRITERIA.

I still have her Christmas cards. There are nine of them.

The denial I keep: Donald Hales, same month, same disease. Fifty eight years old, millwright, forty three pack years of his own choosing and another twenty of somebody else’s. PS 2, not 1. His balance was going, his knees were going, and the oxygen tank had gone back to storage that summer but he still had the marks from the cannula on his ears. His labs were borderline, which is a real category that means: green room, he gets in; red room, he does not. The quarter went red on a Tuesday. The call came down at 10:00. His daughter drove him in at 2:00, because the desk works on hope, and hope does not return calls.

I met them in the intake room. I said, “Mr. Hales, I’m sorry.” He said, “What does the list say?” Which is the sentence I heard a thousand times and never got used to. I told him there was still the trial. He said, “Coin flip.” He said, “Fifty fifty I get it anyway.” He said, “All right. Sign me.” I walked them to the trial desk myself. The trial consent is a different document, and he signed it, and the randomization is a coin with no memory, so I want to be exact about the next part: he drew the control arm.

I stood in the corridor after they left. The trial coordinator said something to me and I did not hear it. I went back to the desk and the phone was ringing and I answered it.

EP 2712. HALES, DONALD. 58. NSCLC, STAGE IV, KRAS G12C. PS 2. DENIED. REASON: PS 2. CAP REACHED. REFERRED TO TRIAL.

The control arm gets the standard of care, which was the taxane and the prayer. Donald Hales died nine months and some weeks later, on the good side of the median, and his months went into the curve, and the curve is what the trial eats, and I fed it with my own hand.

The coin has no memory. Three years later, one of my cannots drew the drug arm and outlived Imani. I have never decided what to do with that fact, so I keep it. Same disease, same desperation, and it lived. The machine built to deny me comfort handed me comfort in exactly one case.

Statisticians do not visit the access desk. Nothing about the meeting was routine. It was year nineteen. The trial was fully enrolled, which should have ended the tension, except the control arm events were still coming in, and events are what a survival analysis eats.

Andre sat in the visitor chair and put a piece of paper in front of me. He had drawn two survival curves. Between them was a third, dashed.

“That one can’t exist,” he said. “It’s the control arm as it would have been if your door had never been there. The counterfactual. No database records it. That’s the point. Nothing records it. Now watch.” He put his finger on the solid control curve. “This is eight point two months. The current best treatment in this population should buy eleven. Why?”

“Because my no’s are sicker than the general population.”

“Your no’s are sicker, poorer, less connected. And the people you admitted, who would have been the healthy end of the randomization pool, they’re over there in the open label cohort, and the company is showing that cohort’s numbers around as a companion study. Seventeen point one months, presented as a mirror of the trial. It’s not a mirror. It’s a mirror of your admission policy.”

He played the paper with his pencil. “The trial is going to claim sixteen point four against eight point two. A doubling, for the headline. The drug’s share of the difference? Maybe six of the months. Maybe seven.”

“The door?”

“Every patient you admitted was a datum withdrawn from the randomization pool. The p value in the charter is conditional on the randomization holding. Randomization only balances the people who showed up to be randomized. It cannot balance your door. The trial will be run correctly. It will be audited forever. And the enrollment was shaped by a desk, and the desk is not a variable, and there is no number in the building that says what I just said to you.”

“There’s the log,” I said.

Andre laughed, which I don’t remember him doing any other time in twenty years. “Right. The pen. The nearest thing to the counterfactual that exists, five thousand entries, in your own hand, and I am the only one in Biometrics who knows what it’s worth.”

He wanted the log. I said no. Policy, and because once I handed it over we would be two people colluding, and I was not ready to be that person. We made a smaller deal. Friday nights, in his office, I read him the names and he did the arithmetic, in his head, writing down only his results, in a hand as small as a man who believes in margins. Andre’s document is twelve pages. It says what I have just told you, in numbers, and it can never be published, because if he is right he is a liability and if he is wrong he is a fool, and he will not sign either verdict.

“Why are you telling me?” I asked him, the first Friday.

“Because when the readout comes,” he said, “somebody has to know what the curves actually say, and know it can’t be said, and keep knowing anyway. A counterfactual needs two people or it erases itself. I can’t be the only one.”

The readout came in April, year twenty. Independent monitoring, then the lock, then the press release, which is how oncology works: the story is written before the analysis is cited. Sixteen point three months against eight point two. A little under a doubling, rounded up in the headline. The p value was under the bar by three orders of magnitude. The charter had been met, and the randomization had held, on paper.

The party was in the atrium. The clinician said something about the first interim read and used the word “clean,” and the table drank to clean data. I drank too. I drank more than I meant to. Clean, I thought. Across the room, Andre watched me watch them. He was drinking water. We did not speak.

Approval came in October, on a Tuesday, faster than anyone predicted. The phrase they use is that the drug sails through, and it sailed. The day it was approved the channel closed, because a drug that exists on the market is not investigational anything, and the mercy clause quietly turned back into a prescription.

The FDA asked for the expanded access records one week later. That part is standard. The request said benefit risk and nothing else. I carried the log to Document Control myself, which made the scan tech nervous, because nobody carries things anymore. His name was Roland. He asked what it was.

“The other side of the ledger,” I said, and he liked that. He opened it at the first page and stood looking at the reason column. “People wrote please,” he said.

“They did.”

He scanned every page. My hand changed as the pages turned: the ballpoint years, the gel years, the flattened grip after the surgery in year seventeen, and at the end, block letters, because block letters are legible, and legibility is the whole job. Roland filed it. The approval stands.

I had already done the one piece of arithmetic I was allowed to do. The week after the readout I asked the stats group for the control roster, and nobody asked why. Twenty years at the desk gets you the file. I matched it at my own desk at night, the log open, the roster pages spread under the best light. Donald Hales was on it, an event, month nine. So were three hundred and ten of my other cannots, the ones I had turned back to the trial when the cap came down. Each one had moved a curve. The curves moved the p value. The p value moved the approval. That is my entire career told as a sum, and I am the only one who ever wrote it down.

The rest of the cannot list, almost six hundred people, never made it into any arm at all. Too sick by the time the door opened, or gone before a referral could catch them. No database holds those people. They are not an event or an arm or a covariate. They exist only in my book, in my first hand. They are not missing from my book. They are missing from everything else.

Brandt came on a Tuesday, the week after Roland scanned the log, retired from the committee the month before, and she did not sit down. She opened her black book on my desk and turned it to face me, holding it with both hands, like paper might get away from her.

“One thousand one hundred and four,” she said. “People who needed the drug before it existed. Who died waiting on a channel that was still being argued about, or who reached the desk on the wrong side of a cap. The cost of the mercy clause. I kept them because I could not do anything else with the names.”

“My log has some of those names.”

“It has the ones who got far enough to apply. Mine has the rest. You held the door. I held the book that stood behind it.”

I did not ask how she had come by the names. I did not need to.

After she left I sat for a while. Then I opened the log to the last page. The final entry was the last admission, a woman from the last program year, a former chemist, PS 1, who sent jam at Christmas. Under it I wrote, in block letters, the reason I had never been allowed to write in twenty years:

THE ONE TRUE REASON FOR EVERY NAME IN THIS BOOK: THE TRIAL NEEDED THEM.

Then I signed it in my first hand, the one that wrote please, and I closed the log and put it back in the cabinet, and the cabinet door still holds, and the approval stands, and the drug works, and the doubling is mostly the drug and partly the door, and I cannot separate them any more than Andre could, and neither can the regulator.

The company is going again. New molecule, same switch, a different floor of the same building. The brochure arrived this morning, and the sentence in it is the same sentence. The open label cohort is already open, the trial is already accruing, the control arm is already hungry, and I am still at the desk. The new log is a binder. The reason column is blank. I have not written in it yet. The old log balances. The new book is waiting, and my hand is steady.