Paired Integers
2029-03-11 07:58 | ward 6N influent, manhole 6N-B | bla_KPC-2, copies/L: 4.1e6
I read the ledger before I read the news, which is the order I have kept for four years and the reason I still have the ledger.
Ten lines of numbers on the left. Ten timestamps on the right. Every dose of ceftazidime-avibactam I have authorized, against the weekly concentration of the gene that defeats it, in the sewage of the ward that produces both. Four years of it now. The spreadsheet is called CONTINUITY. The joke, such as it was, was mine alone.
My working theory, in case anyone ever asks, and nobody has: the two columns are one event seen twice. A course of a reserve agent goes into a patient, most of it excreted unchanged by the kidney, reaches the ward drain within hours, and the drain is where a susceptible population of Klebsiella and E. coli and Enterobacter sits in a biofilm waiting to be recruited. Antibiotic in the lumen selects; antibiotic in the pipe disperses. The gene goes up. On the weeks I hold the line, the number softens. It has softened in eleven of the nineteen quarters since the program started.
I keep the ledger to catch myself. Every signature I regret is on the right, and every rise on the left, and I line them up by hour because hours are what a signature actually spends. It is a confession written in integers. I do not need anyone else to read it.
My theory is falsifiable. That is the part I need on the record before today happens, because today it will need to have been falsifiable and not merely sad.
By nine I have cleared three renewals and declined a fourth, which was a Pseudomonas and not my species, and the fourth call of the morning is Lund, of course.
“The 6N transfer,” he says. “You’ve seen it.”
“I’ve seen the face sheet.”
“Sixty-one, transferred from Ridgeline, fever of thirty-nine point four, lactate is climbing, and I am going to want to open up in about ninety minutes.”
“Open up with what?”
“Whatever you will give me in about ninety minutes.”
I tell him what I tell him every time, which is that I will come to the bedside, and he says the word lactate again, and hangs up, and I go and wash my hands for no reason.
There is a photograph of Ridgeline LTCF taped inside my desk drawer, a two-story brick building in Sycamore with a sign that says Where Every Day Is a Good Day, taken from the county’s last inspection report. The report notes a sink leak in the north wing that was repaired in month three of a fourteen-month deficiency. Long-term care is where the plasmid winters. The hospital is where it is noticed.
The conjugation scene arrives at 14:40, on paper, because Renna prints things.
She is our wastewater epidemiologist, a job title that did not exist when I trained, and she hands me two pages with a pen already lying across them.
“6N,” she says. “Both of them.”
Patient A, bed 614, Klebsiella pneumoniae, carbapenem resistant, KPC-2. Patient B, bed 611, Escherichia coli, carbapenem resistant, isolated from a urine culture the week before. Different species, different patients, different floors of the tree. And on page two, the plasmid fingerprint: IncFIB(pQil-like) replicon, 68.4% average nucleotide identity across the shared backbone, bla_KPC-2 cassette identical to twelve bases.
“Same plasmid,” Renna says.
“Not the same bug.”
“Same plasmid in two bugs in two beds on the same ward, thirty-one meters of pipe between them, and both patients were at Ridgeline within the last year. That is not a coincidence, that is a bus route.”
She tells me the rest while I hold it. Conjugation needs contact, and the contact happens in the gut, in the hours after the transfer, a pilus extended, a copy of the plasmid shut across, and now two organisms that will never share a dinner table share a resistance gene. The pipe is not how it got from bed to bed. The pipe is how it got out of the building. The gut is where it got from patient to patient, and the gut does not care what I sign.
“So the ledger,” I say.
“The ledger is a correlation with your own signature on both ends,” she says, and sits down, and does not soften it. “Can I tell you what it can’t do?”
“Please.”
“It tells you the gene is there. It doesn’t tell you who is carrying it, because you are measuring free and dead DNA in sewage as much as live cells. It doesn’t tell you which plasmid, I did that by sequencing the isolates, not the water. It doesn’t tell you the direction of anything. And it cannot tell you that your dosing moved the number, because there are eleven other variables in that building and you hold a control over none of them.”
“You’re describing my whole career.”
“I’m describing an assay.”
The thing I do not say out loud, the thing that has been sitting on the desk since the first sample: this instrument is exactly the shape of a surveillance program. Influent, effluent, weekly, quantitative, run continuously, with the metadata attached. If someone had wanted early warning of what the ward was shedding, they would have built this. What I built instead was a confessional with a qPCR machine attached, and I have been reading my own guilt into it for four years.
That is the correction that should have ended it, and instead it starts it.
Because Renna’s argument is about attribution, and attribution is not the only thing a number can do. A number can also be a trend.
After she goes, I pull the second tab of the spreadsheet, the one I opened in year two and have checked perhaps forty times, and I read it upward.
2029-03-12 07:58 | municipal intake, above hospital discharge | bla_KPC-2, copies/L: 3.7e6
The intake sample is the other column. The city draws its drinking water from the river fourteen hundred meters upstream of the hospital’s outfall, and the lab runs the same assay on it, because a colleague of Renna’s does it for a grant and shares the file, and I have been pasting it beside my own numbers since 2025.
Reading upward means ignoring my signatures. Ignore the timestamps on the right and look only at the left column, week over week, four years, and then read the intake beside it.
The intake was 3.1e6 in the first year. It is 3.7e6 now. It never went down. It did not soften in the quarters I held the line, and it did not spike in the quarters I signed. It rose at a slope that does not know the difference between my two columns, and it rose while the stewardship program was doing the most aggressive work in the hospital’s history, and it rose while I was at home in the evenings believing I was bending something.
The gene was in the municipal water before my program, or at least without reference to it. Every dose I withheld did not move it. Every dose I signed did not move it. My withholding and my signing are, against that baseline, indistinguishable from nothing.
I sit with it for a long time. The honest reading is that I have four years of evidence that I was not a variable. The dishonest reading, the one that wants in immediately, is that I have four years of evidence that I was innocent. And the second reading is wrong for a reason I have to write down in order to stop myself believing it: absence of my effect is not absence of the disease. The river does not care about my theory, and it does not acquit me, it just was never listening.
Also, and Renna would say it before I finished the thought: I do not have the flow. I have copies per liter at a manhole that runs twelve cubic meters a day and copies per liter in a municipal influent at a treatment plant handling something on the order of a million cubic meters a day. I have been comparing concentrations across two flows that differ by five orders of magnitude. The right unit was always load, copies per day, and I never once converted. The hospital’s whole output, if I integrated it, is a rounding error in the intake. Which means my upward read, the one that just took my theory apart, was itself done with a ruler I never calibrated.
That is the instrument. It answered neither question. It answered neither the question I asked it, which was whether I was killing people, nor the question it could have answered, which was whether the ward was shedding.
2029-03-13 04:12 | ward 6N, bed 614 | speciation pending, carbapenem MIC > 16
Lund is alone in the room when I get there, and the family is not, which means the family is in the room with him.
There are four of them. A daughter holding the phone in both hands. A son-in-law against the wall. An older woman in a chair that has been pulled close to the bed, and she is the one talking, and what she is saying is he has been waiting for the test, over and over, in a voice with no grammar left in it.
Lund does not look up. He is reading the monitor and doing arithmetic on the back of a chart.
“He’s holding,” he says. “MAP is fifty-eight on norepinephrine, lactate is four point one and it was three point two at midnight, and I have a gram of meropenem running that will do nothing at all, because the rapid panel came back carbapenem resistant forty minutes ago.”
“Then we change.”
“Then we change, and the question is to what, and how fast you will hand it to me.”
I tell him what I have. Ceftazidime-avibactam, six grams loading, then two grams every eight hours, and I can have it at the bedside inside two hours if I sign now. And he does the thing he does, which is to hold still and let me finish, and then say the number.
“Kumar, 2006,” he says. “Seven point two percent mortality if you get effective antimicrobials in the first hour of shock. That’s the number. It goes up seven point six percent an hour after that, every hour, and by hour six you are in the forties. I have been reciting it to residents since before you were on the P&T. The families do not care about the plasmid. The plasmid has never once looked at me at four in the morning.”
“The resistance horizon this program defends is longer than the lifespan of anyone in this unit,” I say, because it is my line, and it is true, and I hear how it sounds in a room with a woman saying he has been waiting for the test.
“Yes,” he says. “It is. And I am the man who has to be wrong in the right direction for the next four hours.”
He is not refusing the data. He has read the data. He read it when I sent it, and this is the part I had braced for and got anyway: he grants it, cleanly, in the next sentence.
“Your importation argument holds,” he says. “Both of them came off that bus from Ridgeline. You want to say the LTCF brought it in and not my prescribing, fine, I’ll say it in the meeting. That does not change the arithmetic in this room.”
The arithmetic in this room is one patient, four hours, and a family that has been waiting for a test for six days. He is doing triage. I have spent four years doing cosmology. And the worst of it is that he is right about the schedule. He is not wrong about anything tonight. He is standing at the edge of a horizon that will not arrive in his career or mine, holding a number that arrives in minutes, and he would be a fool to trade the minutes for the horizon.
The resident is looking at me. The daughter has stopped saying the sentence and started listening.
“The dose is two grams every eight hours,” I say. “The window is shortest possible. If the isolate comes back pan-susceptible I pull it in twenty-four hours.”
“Twenty-four,” Lund says, and writes it on the chart, and does not thank me, which is correct.
2029-03-13 06:41 | authorization 4119 | ceftazidime-avibactam, 6 g load, 2 g q8h, 24 h window, expires 2029-03-14 06:41
The act is easy. It takes forty seconds. The dose goes through, the pharmacy pump is programmed, and I stand in the corridor outside 614 with the phone still in my hand and understand, without any language for it yet, that I have just done the exact thing my ledger exists to prevent, and that the ledger, read upward, says the exact thing my ledger exists to prevent is irrelevant.
Restraint and ruin are the same action here. Not as a paradox. As a fact about two different timelines that I am now on both sides of.
I go back to my office and open CONTINUITY, and the last line I enter is the one above, timestamped with the authorization number, because a ledger that stops before the moment it mattered is not a ledger.
Then I open the report file. The quarterly, which I chair, which is due Friday, which will be read by six people who have not seen a spreadsheet and two who have.
The first sentence takes four tries. The fourth one is the truth, and it is four clauses long, and I leave it standing on the page and do not add the paragraph that would soften it, because the softening paragraph is the part I have been writing for four years.
Between 2025 and 2029, the hospital’s own wastewater surveillance detected a sustained increase in bla_KPC-2 abundance upstream of its discharge, a trend the stewardship program neither produced nor interrupted, and which cannot be attributed to hospital prescribing on the data available.
That is what the instrument says. It says the trend is there, that we did not make it, and that we cannot say who did. Three claims, all of them defensible, none of them an acquittal, and I know exactly how the committee will read the middle clause, which is as an exoneration, which is not what it says and not what I meant and which I will not correct, because the correction would require me to explain that a program can be pointless without being blameless, and I do not have the word for that yet.
I put the cursor at the end of the sentence and wait for the second one.
2029-03-13 06:41 | ward 6N influent, manhole 6N-B | bla_KPC-2, copies/L: 4.1e6 | authorization 4119