Deep divefrom Episode 12Fat loss11 November 202619 min

Grip, Gait and the Road to Diabetes

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Beyond functional disability: joint trajectories of muscle strength and walking speed and new-onset diabetes risk (Frontiers in Endocrinology, 2026)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science, where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 2Today is one paper, published on October twenty-eighth in Frontiers in Endocrinology by three researchers at the First Affiliated Hospital of Wenzhou Medical University in China. It asks whether the way grip strength and walking speed change over several years says something about the chance of developing diabetes, in older people who still manage everyday life without help.

Host 1So the people in this study are not frail or disabled, they are older adults who can still dress, wash and get around on their own?

Host 2Exactly, that is the group. Anyone who had trouble with any of six basic daily activities was left out, and so was anyone who already had diabetes.

Host 1And what is the short version of what they actually found in those people?

Host 2In four large ageing surveys, from the United States, England, continental Europe and China, a small group whose strength and walking followed an unfavourable path was diagnosed with diabetes at roughly two and a half to four times the rate of everyone else. It is an association in observational data, and there are real questions about how the groups were defined, which we will get to.

Why a handshake became a health test

Host 1Before we get into it, why would anyone think grip strength has anything to do with diabetes in the first place?

Host 2The wider picture in physiology is that muscle is not only for moving, it is the body's biggest user of blood sugar. After a meal, most of the sugar that insulin clears from the blood goes into skeletal muscle, so if muscle shrinks or works less well, the body handles sugar worse.

Host 1How did squeezing a handle and timing a short walk become things doctors care about at all?

Host 2Over the past few decades, large studies found that a weak grip predicts earlier death and more hospital stays, and walking speed came to be called a kind of vital sign in medicine for older people. Both tests are cheap, take a minute, and need only a handheld gauge called a dynamometer and a stopwatch.

Host 1And that checklist of daily activities you mentioned, where does it come from?

Host 2It goes back to the Katz index from the nineteen sixties, a list of basics a person either can or cannot do without help. Here they used six items shared by all four surveys: dressing, bathing, eating, getting in and out of bed, using the toilet and walking indoors.

What was known before

Host 1So what did researchers already know about strength, walking and diabetes before this paper came along?

Host 2The authors cite earlier work showing that lower grip strength and slower walking, each measured once, go with a higher risk of diabetes later. The muscle and sugar link makes that biologically plausible.

Host 1If that is already known, what exactly was still missing from the picture?

Host 2The authors point to three gaps. Most studies measured function only once, most looked at strength or walking separately, and most used a single group of people from one country.

Host 1Why does measuring just once matter, if a weak grip is simply a weak grip?

Host 2Because one measurement cannot tell apart a person who has always been on the weaker side from a person who is losing strength quickly. Following people over several visits shows the direction and speed of change, not only the starting point.

The open question

Host 1So the question here is whether the path over time, strength and walking together, tells you more than a single snapshot does?

Host 2That is it, with one extra condition. They wanted to know whether this works in people who show no outward disability yet, the stage where an early warning would actually be useful.

How they did it

Host 1Where did the data come from, did they recruit their own volunteers for this?

Host 2No, they reused four existing national ageing surveys: the Health and Retirement Study in the United States, the English Longitudinal Study of Ageing, the Survey of Health, Ageing and Retirement in Europe, and the China Health and Retirement Longitudinal Study. These surveys interview and examine the same people roughly every two years.

Host 1How many people ended up in the analysis in the end?

Host 2Two thousand six hundred in the American survey, three thousand and forty-three in the English one, one thousand five hundred and thirty-three in the European one and three thousand five hundred and fifty in the Chinese one. That is about ten thousand seven hundred people, analysed as four separate studies rather than one pool.

Host 1Those sound like decent numbers, but how many people did they start out with?

Host 2Far more, and that matters. The European survey started with thirty thousand four hundred and sixteen people and kept about one in twenty, mostly because twenty-one thousand and thirty-nine had no walking measurements, and the American one went from eighteen thousand four hundred and sixty-nine down to two thousand six hundred.

Host 1Why would so many people be missing a simple walking test?

Host 2The paper does not say, but as far as we understand the European survey's design, it only timed walks in its oldest participants in those early rounds, which fits its median age of seventy-eight. More generally, you needed three complete rounds of both tests with no disability and no diabetes, which favours people who stayed well and kept turning up.

Host 1How exactly did they measure the two things, strength and walking?

Host 2Walking speed was a short timed walk at normal pace, eight feet in the three Western surveys and two and a half metres in China, done twice and averaged. For grip, both hands were tested on a dynamometer, and the best result from each hand was added into one total in kilograms.

Host 1Eight feet is not even three metres, so is that really a walking test?

Host 2It is very short, and that is standard in these surveys because it can be done in someone's living room. It captures how briskly a person sets off and moves, not their stamina.

Host 1Then how do you turn three rounds of measurements into something called a trajectory?

Host 2They used group-based trajectory modelling. It assumes the population is a mix of a few hidden groups, each with its own typical curve over time, and it estimates those curves and which group each person most likely belongs to.

Host 1So the computer sorts people into a handful of typical paths, a bit like sorting runners by how their pace changes during a race?

Host 2That is a fair picture. Here it sorted on strength and walking jointly, tried between one and four groups, and picked a solution using a fit score called the Bayesian information criterion, with rules that every group hold more than five percent of people and that people be assigned with at least seventy percent confidence on average.

Host 1And how many groups came out of that sorting in the end?

Host 2Two in every cohort. In the three Western surveys they are called high-level decline and low-level stability, the reference group, versus low-level decline and low-level rise; in China, high-level decline and low-level decline versus low-level decline and high-level decline.

Host 1I have to admit those names lose me, which part refers to strength and which part to walking?

Host 2The paper does not spell that out, and the labels are hard to decode even on the page. The authors describe the higher-risk paths as persistently low function or progressive decline, and the figure labels suggest that group was small, about five percent of people in at least two of the cohorts.

Host 1Once people are sorted into groups, how do they get from there to diabetes?

Host 2They followed everyone from the end of those three rounds until a first diabetes diagnosis, death, dropping out or the end of the study. Diabetes meant a doctor's diagnosis that the person reported in the interview, with no blood tests.

Host 1How long were people followed after that, and how many got diabetes?

Host 2The median follow-up was ten years in the American survey, twelve in the English, six in the European and eight in the Chinese. Over that time there were three hundred and forty-one, two hundred and sixty-five, one hundred and seventy and four hundred and forty-eight new diagnoses.

Host 1Older people with weak muscles probably differ in lots of ways, heavier, poorer, less active, so how did they stop those differences from fooling them?

Host 2They accounted for up to fifteen factors, including age, sex, education, wealth, body mass index, high blood pressure, smoking, drinking, heart disease, cancer, hip fracture, physical activity, memory and mood. They did it two ways at once: a weighting method called inverse probability of treatment weighting, plus ordinary statistical adjustment on top.

Host 1What does that weighting actually do, if you put it in plain words?

Host 2It gives each person a weight so the two groups end up looking alike on those factors, as if they had been balanced by design. Combining it with ordinary adjustment is called doubly robust, because if one of the two is a bit off, the answer can still hold.

A word from our sponsor

Host 1This episode is brought to you by Downslope, the indoor corridor that tilts gently downhill in whichever direction you walk.

Host 2Unlike certain fold-out hills, Downslope never asks for extra oxygen. It only gives.

Host 1Every Downslope is exactly eight feet long, the length of a standard usual-pace walking test. It is propensity weighted, truncated at both ends, and doubly robust.

Host 2I installed two. Over three waves I moved from low-level decline to high-level rise, and my grip strength now sits in a group of its own.

Host 1Placed end to end, two Downslopes form a small valley. In our longest follow-up, one customer walked downhill for twelve years and arrived before she left.

What they found

Host 1So, now that we know how it was done, what were the actual numbers?

Host 2In the fully adjusted model, the less favourable group had a hazard ratio of four point one seven in the American survey, three point eight five in the English, three point three eight in the European and two point six in the Chinese. Every one was clearly above one, with the lowest end of any ninety-five percent confidence range at about one point nine.

Host 1What does a hazard ratio of four mean for a real person, are they four times as likely to get diabetes?

Host 2Roughly, it means that at any point during follow-up, people in that group were being diagnosed at about four times the rate of the reference group. It is a relative number, so on its own it says nothing about how many people that is.

Host 1Do we know the absolute numbers then, how many people actually got diabetes in each group?

Host 2Across each whole cohort, about nine to thirteen percent were diagnosed during follow-up, for example three hundred and forty-one of two thousand six hundred in the American survey. The main text does not give the rate within each trajectory group, so you cannot read off an absolute risk for the small high-risk group.

Host 1Did the numbers change much between a raw comparison and the fully adjusted one?

Host 2Very little. In the American survey the crude hazard ratio was four point six eight and the fully adjusted one four point one seven, so the measured factors explained only a small part of the gap.

Host 1Is that reassuring, or is it something that should make us suspicious?

Host 2It can be read either way. It suggests the measured factors were not driving the link, but it cannot rule out things they did not measure, such as blood sugar that was already creeping up.

Host 1Was it really the combination that mattered, or would grip strength alone have done the job?

Host 2They checked that, and paths built on strength alone or walking alone showed limited and inconsistent links with diabetes. The joint paths were more consistent, though those single-measure results sit in the supplement, not the main paper.

Host 1What about people who died before they ever had the chance to get diabetes, does that skew things?

Host 2It could, so they ran a competing risk model that treats death as a separate outcome. The estimates barely moved: three point eight eight, three point three seven, three point one eight and two point four seven.

Host 1Did they try any other way of handling the timing of diagnoses?

Host 2Yes, a model that treats each two-year survey round as one step, which gives odds ratios instead. Those came out larger, between about four and five in every cohort, including four point eight eight in China, nearly double the main estimate there, and the paper does not discuss why.

Host 1And did the link hold across men and women, and across younger and older people?

Host 2Mostly yes, across age, sex, weight, blood pressure and activity level. The one soft spot was the European survey, where people seventy or younger had a hazard ratio of two point two two with a range that included no effect, probably because that survey had few younger people.

Host 1The summary also mentions machine learning, so what did the computers add to all this?

Host 2They built nine standard prediction algorithms plus a hundred and twenty combinations of them, a hundred and twenty-nine models per cohort, to predict who would get diabetes. Then a method called SHAP showed how much each input, including the trajectory group, pushed each prediction up or down.

Host 1So did adding the trajectories make the predictions noticeably better than without them?

Host 2The paper does not really answer that. It reports which combinations ranked best and says the trajectory feature contributed, but the main text gives no headline accuracy figure, and as far as we can read the chart for the American survey, the trajectory counted for less than high blood pressure or age.

Host 1Did they test the models on a population the models had never seen before?

Host 2Once: they trained on the American and English surveys, tuned on the European one and tested on the Chinese one. A different algorithm came out on top in that test, which mainly tells you the best method depends on the population.

Why it matters

Host 1Stepping back, what is the bigger idea here, beyond these four surveys?

Host 2The idea is that physical function is a moving signal, not a fixed trait, and that strength and walking declining together might flag metabolic trouble before any disability shows. It fits the view of muscle as a metabolic organ, not just machinery.

Host 1Who would actually use this in practice, a family doctor during a check-up?

Host 2That is what the authors suggest: repeating a grip test and a short timed walk at check-ups as a cheap extra, so that someone sliding on both gets their blood sugar looked at more closely. They say explicitly that this should add to standard diabetes screening, not replace it.

Host 1So if my own grip is getting weaker, does that mean I am heading for diabetes?

Host 2No. This is a group-level association, most people in every group were not diagnosed during follow-up, and a blood test remains the way to find out.

Host 1Does it at least show that getting stronger would protect someone from diabetes?

Host 2It does not, because nobody was given any intervention. The wider picture from trials is that exercise and weight loss lower diabetes risk in people at high risk, but this paper cannot say whether improving a trajectory changes anything.

Honest caveats

Host 1What are the biggest reasons to be careful with this result?

Host 2First, the outcome itself. Diabetes here is a diagnosis people reported, with no blood sugar test, so undiagnosed cases are missed, and people who see doctors more often get diagnosed more often.

Host 1Could early diabetes be causing the decline, rather than the other way round?

Host 2That is a real possibility the design only partly addresses. They excluded people diagnosed during the three rounds, but blood sugar can rise for years before a diagnosis, and that hidden phase could itself weaken muscles and slow walking.

Host 1What about all the people who were dropped, does that change who the results apply to?

Host 2It does, and the authors say so. Those included differed from those excluded on age, mood, weight, wealth and some illnesses, so the results describe people healthy and steady enough to complete three rounds of tests.

Host 1Is there anything to worry about in how the groups themselves were defined?

Host 2Several things. The groups were found separately in each cohort, so the high-risk group in China is not the same kind of group as the one in England, which is why the authors did not pool the results; and with one group that small, a few hundred people carry most of the signal.

Host 1Should we worry about how carefully the paper itself was put together?

Host 2There are editing slips. In places the text calls an odds ratio of four point three a three point three fold increase, or a ratio of three point one eight a two point one eight fold hazard; the tables are what count, but it calls for careful reading.

Host 1And the machine learning part, how much weight should a listener give that?

Host 2Not much yet. The data were split at random rather than by time, which the authors acknowledge, the rare diabetes cases were boosted with synthetic examples during training, and there is no clear figure for what the trajectories add on top of the usual risk factors.

Host 1One thing I wonder about, were the people who got diabetes mostly the older ones?

Host 2No, in three of the four surveys they were younger than those who stayed free of it, and they had higher body mass index. It is a reminder that weight remains a major driver, and that in older groups, who stays in a study long enough to be diagnosed is a particular mix.

What to watch next

Host 1What kind of study would convince you that this is a genuine early warning sign?

Host 2One that also measures blood sugar at every visit, so hidden diabetes can be excluded and the trajectories checked against glucose directly. The authors also call for pooling the original data from many cohorts using trajectory groups defined in advance, the same way everywhere.

Host 1And before doctors use it, what would a fair test of it look like?

Host 2A prospective test, meaning people are classified by their trajectory now and followed to see whether the predictions hold. The number to watch is how much grip and walking trends improve prediction beyond age, weight and blood pressure.

Sign-off

Host 1So, to sum up, what should someone take away from this paper?

Host 2That in four ageing surveys on three continents, a small group whose strength and walking followed a poor path over a few years was later diagnosed with diabetes about two and a half to four times as often. It is a consistent association, not proof of cause, and not yet a screening tool.

Host 1That's the deep dive. Every weekend this same feed carries the roundup, the week's new papers in one short episode, timed over a standard eight feet.

Host 2Thanks for listening to the Hastrup Review of Science.