Cancer Wasting Was Hiding in Two Million Routine Weigh-Ins

Tobias Janowitz’s mother grew up in a childhood “shaped by war, malnutrition, and displacement.” Days before she died of cancer, she told her son: “My body is all used up, and I have no will left to live.” Janowitz, a physician-scientist at Cold Spring Harbor Laboratory, later opened an essay in Neuron with that sentence and used it to argue that what patients report about their own bodies belongs in the dataset, not in the anecdote pile. The condition that killed her, cancer cachexia, had been leaving a record of its own the whole time, one routine weigh-in after another, in the files of the hospitals treating it.

Ed Reznik, a data scientist at Memorial Sloan Kettering, heard Janowitz describe a patient’s last words at a seminar and only afterward learned whose they were. “It turned out to be his own mother, who died of cancer-related cachexia,” Reznik told MSK’s news office. Janowitz is a coauthor on the cachexia atlas MSK published this week in Nature Medicine. The team built it without ordering a single new test, because every measurement in it had already been taken during ordinary care.

Two million weigh-ins

Cachexia is the wasting of advanced illness, a loss of weight and skeletal muscle that extra food doesn’t fully reverse. The 2011 international consensus definition offers several routes to the diagnosis. MSK’s team used the central one, a loss of at least 5 percent of body weight within six months, and searched the weight history of every genetically sequenced patient for any window that crossed it. The preprint posted in September 2025, which carries the full tables, covered 59,493 patients and 2,335,098 BMI measurements, roughly 39 per patient over an average of about five years. The version published this week reports a somewhat smaller cohort of 52,236.

Measured that way, 48.8 percent of the preprint cohort had at least one cachectic episode at some point during follow-up. At stricter cutoffs, 23 percent lost 10 percent of their weight and 9 percent lost 15 percent. The algorithm logged 37,544 separate episodes, and 24.1 percent of the patients who had one went on to have another.

That last figure doesn’t fit the way cachexia is usually taught, as a one-way slide at the end of life. For some patients that is exactly what it is, but the records also show bouts that came and went: the weight fell, sometimes came back, and in about a quarter of the affected patients fell again. “Cachexia wasn’t always just a long, steady decline,” said co-lead author Amy Xie. By the preprint’s count, a typical patient spent about 11 percent of follow-up inside an active episode.

An albumin level that passed

Low serum albumin is the textbook lab sign of a wasting patient, and in this cohort it turned out to be a weak signal. In colorectal cancer, 33.7 percent of albumin readings taken during an episode fell below the study’s 3.4 g/dL cutoff, against 12.8 percent of readings taken outside one. That leaves roughly two of every three readings from a patient who was actively losing weight sitting above the line. The declines, Xie said, “might also be subtle enough not to raise a red flag unless clinicians are actively looking for them.”

The scale caught every one of those episodes. The diagnosis list, in at least one large hospital, mostly did not. At Rouen University Hospital in France, a team that included cachexia researcher Vickie Baracos searched six years of records, 2018 through 2023, covering 76,547 cancer patients. They found a cachexia diagnosis in 2.42 percent, and only 620 patients, or 0.81 percent, carried R64, the ICD code for it. Among patients with colorectal, pancreatic or lung cancer alone, another 1,340 had a BMI under 20 or weight loss over 5 percent in their structured data and were never coded as cachectic. Code searches alone, the authors concluded, “underestimated cachexia prevalence,” which is a polite way to describe a miss that large. A missing code doesn’t prove a patient went without a dietitian. It does mean that nobody can easily track, audit or study a diagnosis that never makes it into the chart.

The survival numbers are stark, and they need careful reading. In the preprint’s unadjusted time-dependent models, an active episode was associated with a hazard of death 14.39 times higher in prostate cancer, 9.64 times higher in invasive ductal breast cancer, 5.54 times higher in colorectal cancer and 2.29 times higher in pancreatic cancer, and the association held in all 19 cancer types the team examined. Some of that runs in reverse: people who are dying lose weight, so part of the signal is death showing up on the scale early. The authors write that their findings “establish associations rather than causation.” The progression data point the same way, more modestly. In colorectal cancer, 31 percent of progression events fell inside a cachectic episode, against the 23 percent expected by chance, and in pancreatic cancer it was 44 percent against 38 percent. Plenty of episodes came with no recorded progression at all.

The tumor seems to have a say too. TP53 mutations tracked with higher cachexia risk in nine cancer types. In lung adenocarcinoma, STK11 carried an adjusted hazard ratio of 1.42, and in colorectal cancer SMAD4 came in at 1.24 after adjustment. Using genotype, stage, performance status and baseline weight, the team built risk models for lung and colorectal cancer that MSK calls “moderately accurate.” That’s fair. Concordance indexes of 0.642 and 0.649 mean that when the model compares two patients, it picks the one who will start wasting first about 65 percent of the time, against 50 for a coin, and so far it has only been tested on a held-out half of MSK’s own patients.

Weight at week 12

Reznik wants the scores to lead somewhere larger. “That’s really the dream, a precision trial for cachexia,” he said. “I don’t think we’re that far off.” Drug development is further along and aimed at a different patient. No drug is approved for cancer cachexia. Pfizer’s anti-GDF-15 antibody ponsegromab produced weight gain in a 187-patient phase 2 trial published in NEJM in 2024. The company is now recruiting for a phase 2b/3 study planned at 982 patients with metastatic pancreatic cancer, and every one of them must already meet the weight-loss criteria for cachexia to enroll. Its primary endpoints are percent change in body weight and an anorexia-symptom score at 12 weeks, with primary completion estimated for January 2028.

The industry is paying to measure weight at week 12 in patients who are already wasting. The atlas was paid for with public and philanthropic money, with the National Cancer Institute, the Department of Defense and Cancer Grand Challenges among the funders, and the most useful thing it offers may be something nobody can sell. MSK’s own write-up says catching cachexia earlier could bring “additional nutritional and rehabilitation supports,” which in practice means a dietitian, a strength program and closer attention to a patient whose weight is slipping. None of that needs a patent or a 2028 readout. It needs someone to look at the weights the clinic has already recorded.

Janowitz’s mother described the whole syndrome in one sentence. Over six years and 76,547 cancer patients, Rouen University Hospital entered the code for it 620 times.

Sources

1. Nature Medicine: MSK’s episode-resolved atlas of cancer cachexia (2026)
2. Nature Medicine research briefing: “Characterizing the onset and progression of cancer cachexia using routine clinical data” (2026)
3. MSK cachexia atlas preprint (2025)
4. Colardelle et al., Journal of Cachexia, Sarcopenia and Muscle (2026)
5. Memorial Sloan Kettering: “Insights About Cachexia, One of Cancer’s Most Common Complications, Hiding in Routine Patient Data”
6. ClinicalTrials.gov: NCT06989437, Pfizer ponsegromab phase 2b/3 in pancreatic cancer cachexia
7. Pharmaphorum: ponsegromab phase 2 results at ESMO 2024
8. Cold Spring Harbor Laboratory: “Patient insights inspire discovery,” on Janowitz’s Neuron essay (2025)