Exercise Non-Responders: Why One Retest Doesn’t Make One

By Marc Ferrer · Founder, Engram Kinetics · Published September 2026 · Updated October 2026

Short answer. A flat retest doesn’t make a client a non-responder. One pre/post comparison can’t separate a real lack of adaptation from the test’s own noise, and a laboratory VO2max test already carries a day-to-day error of roughly 3 to 6% in the same person. When trials raise the training dose, the share of people classified as non-responders falls, in two well-known trials all the way to zero. In studies that measured several outcomes, nearly everyone improved on at least one. Real differences in trainability probably exist, but one retest can’t show you who has them. Check the test and the dose, look at the outcomes that matter to the client, and adjust the program before you label the person.

Twelve weeks in, your client’s retest comes back flat. VO2max went from 30.1 to 30.4 mL/kg/min. Attendance was good and the program looked sensible on paper. A word is waiting for you: non-responder.

It’s a comfortable word. It sounds like physiology and it has a famous study behind it. It also moves the problem out of your program and into your client’s genes, and it closes the case. That’s where it goes wrong. Over the last decade, research on individual responses has turned against the way the label gets used in practice. Here is what it shows, and what to check instead.

Where the idea comes from

The non-responder has a respectable origin. Bouchard’s analysis of the HERITAGE Family Study covered 481 sedentary adults from 98 families who trained on cycle ergometers for 20 weeks. VO2max was measured twice before training and twice after. The average gain was about 400 mL/min, but individual gains ranged from little or nothing to more than 1 L/min. The response also ran in families, with a maximal heritability estimate of 47% (Bouchard et al., 1999).

Those results put trainability into the textbooks, and ACSM’s Guidelines reflect them. The chapter on exercise prescription in the 12th edition warns that some people may not respond as expected, given the considerable individual variability in response to the same regimen. The sentence is accurate. Read two of the studies it cites, though, and the practical message shifts. In one, the share of non-responders fell as the weekly dose rose (Sisson et al., 2009). In the other, every participant improved on at least one of the variables measured (Scharhag-Rosenberger et al., 2012). Non-response depended on the dose in the first and on the outcome in the second.

What a single retest can’t see

Every measurement carries noise, and for VO2max much of it comes from the person rather than the equipment. Katch and colleagues retested five trained subjects 8 to 20 times each over two to four weeks. Their day-to-day variability, expressed as a coefficient of variation, came to about 5.6%, and 90% or more of it was biological (Katch et al., 1982). In HERITAGE, the two-day test-retest coefficient of variation was 5.0% before training and 4.1% after (Shephard et al., 2004). Other laboratories report similar or smaller values, such as 4.9% (Edgett et al., 2018) and 2.8% (Zinner et al., 2023).

Now compare that with what you’re trying to detect. A change between two tests carries the error of both, so its noise is about 1.4 times the test’s typical error. If nothing real had changed, about 84% of retests would still land within twice the typical error of the first result, and 95% within about 2.8 times. So a change has to pass roughly 2.8 times the typical error before you can call it unlikely to be test noise by the usual 95% standard. With a typical error of 3 to 5%, that bar sits at roughly 8 to 14% of the starting value. A client who truly gained 5% can produce a flat retest, and a client who truly gained nothing can produce a 5% improvement. If the VO2max was estimated from a submaximal or field test instead of measured, the error is larger still (see direct versus estimated VO2max).

Statisticians have been making this point for years. Atkinson and Batterham showed with simulated data that the familiar chart of individual pre/post changes, sorted from worst to best, can suggest large individual differences even when every simulated person has exactly the same true response. Random day-to-day variation and regression to the mean are enough to produce it (Atkinson & Batterham, 2015). To estimate true individual differences, you have to compare the spread of changes in a group that trained with the spread in a group that didn’t.

When researchers make that comparison, the true spread shrinks. A 2024 meta-analysis of 24 aerobic training studies with non-exercising control groups concluded that most of the variation in observed VO2max changes was due to measurement error, a term that here covers the person’s ordinary day-to-day variation as well as the equipment’s. It found no strong evidence of true differences in VO2max trainability within single programs (Renwick et al., 2024). An analysis of individual data from 1,879 participants in eight randomized trials reached a similar conclusion for fitness, waist circumference and body mass: the variability in observed changes likely reflected measurement error and behavior rather than differences in trainability (Bonafiglia et al., 2022).

Even in HERITAGE, noise accounts for part of the spread. Shephard’s reanalysis estimated that allowing for test-retest error cuts the standard deviation of individual responses from 202 mL/min to roughly 123 to 138 mL/min, while substantial differences remain (Shephard et al., 2004). So the honest summary isn’t that everyone responds alike. It’s that most of the scatter you see in pre/post changes is noise, with a smaller true spread underneath. Keep in mind that these are group-level analyses. Under some assumptions, they estimate how much extra spread training adds across a population. They can’t tell you where one client sits in it.

Raise the dose and the non-responders thin out

Whatever non-response is, the dose changes how often you see it.

Sisson’s analysis of the DREW trial covered 310 sedentary postmenopausal women who trained for six months at 50% of VO2max, at one of three weekly doses. Counting as a non-responder anyone whose VO2max didn’t rise at all, the share fell from 44.9% at the lowest dose to 23.8% and then 19.3% at the higher ones (Sisson et al., 2009).

Ross and colleagues separated amount from intensity in 121 sedentary adults with abdominal obesity, all of whom completed at least 90% of five weekly sessions over 24 weeks. A low amount at low intensity left 38.5% of participants as non-responders. A higher amount at the same intensity cut that to 17.6%. The same higher amount at 75% of VO2peak left none, 0 of 31 (Ross et al., 2015).

Montero and Lundby went further. Seventy-eight healthy adults trained for six weeks with one to five 60-minute sessions a week. With non-response defined as a change in peak power output within the test’s typical error, they counted 69%, 40%, 29%, 0% and 0% non-responders across the five groups. The non-responders then trained six more weeks with two extra sessions a week, and every one of them responded (Montero & Lundby, 2017). Two cautions apply. The study had no non-exercising control group, which critics pointed out. And people picked out for a low first change tend to show a larger second one anyway, through regression to the mean. The direction still matches the controlled trials above.

Intensity appears to matter as much as volume. In another trial, 16 of 31 untrained adults were VO2max responders after 10 weeks of moderate training. The group was then split in two, balanced for that first response. The half that stayed at moderate intensity for 16 more weeks produced no new responders. The half that moved to higher intensity and then to interval training, at the same energy expenditure, ended with 13 of 15 responders (Reuter et al., 2023).

There’s a statistical catch. Counts of responders depend on the group’s average change: raise the average and fewer people fall under any fixed threshold, even if nobody’s biology differs from anyone else’s (Atkinson et al., 2019). So these counts don’t prove that a higher dose rescued anyone in particular. What they do show is practical: the dose is a variable you control, and it’s the first one to revisit.

Non-response is usually about one outcome

The label usually rests on one number, most often VO2max. In the studies that measured several outcomes, very few people failed on all of them.

In a one-year program, 4 of 18 adults showed no VO2max gain, yet every participant improved on at least one of VO2max, resting heart rate, exercise heart rate or the individual anaerobic threshold (Scharhag-Rosenberger et al., 2012). In a crossover study of endurance and sprint interval training, all 21 adults responded on at least one variable (Bonafiglia et al., 2016). After four weeks of training, 13 of 25 men didn’t improve VO2peak beyond measurement error, and all 13 improved at least one submaximal measure (Clarke et al., 2017).

Resistance training in older adults shows the same pattern. Among 110 people over 65, every one improved on at least one of lean mass, muscle fiber size, leg strength or chair-rise time. The authors titled their paper “There are no nonresponders to resistance-type exercise training in older men and women” (Churchward-Venne et al., 2015). And responsiveness doesn’t cluster in the same people. When HERITAGE ranked 564 participants by their response on seven cardiometabolic traits, only one landed in the most responsive or the least responsive fifth on all seven (Barber et al., 2022). One review summed it up: global non-responders to exercise are unlikely to exist (Pickering & Kiely, 2019).

Those counts need one caveat. The more outcomes a study measures, the more likely at least one improves by chance, so they describe a pattern rather than prove that everyone adapted.

The label doesn’t even repeat

If being a non-responder were a stable trait, the same person should respond in much the same way to the same program twice. Two studies checked. Fourteen men did two identical four-week interval blocks three months apart, and the ranking of their individual VO2max responses was poorly reproduced, with an intraclass correlation of 0.37 (Del Giudice et al., 2020). Forty-two untrained adults did two identical eight-week endurance blocks separated by eight weeks of detraining. The reproducibility of each person’s VO2max response was poor, with an intraclass correlation of 0.19 (Odden et al., 2026).

That doesn’t prove trainability is pure noise, since the same measurement error that blurs one retest also drags these correlations down. It does mean that one training block tells you little about a person’s trainability.

Where the skeptics have a point

None of this means everyone adapts alike. The HERITAGE family data are real, and a 2019 consensus statement on exercise response variability concluded that the same exercise dose produces variable responses, with a genetic contribution (Ross et al., 2019). A 2025 individual-participant meta-analysis found a larger spread of VO2max changes in exercise groups than in control groups, 1.9 versus 1.3 mL/kg/min, which points to some true heterogeneity (Meyler et al., 2025). After six weeks of very low-volume sprint training, one pooled analysis estimated that 18% of participants were likely non-responders (Metcalfe & Vollaard, 2021). And not every non-responder converts. Of 22 older women whose muscle mass gain stayed within measurement error after 12 weeks of resistance training, only 3 crossed the threshold after 12 more weeks at a higher weekly volume (Nunes et al., 2020).

So some people probably gain less from a given program than others. The claim here is narrower. With one client and one retest, you can’t tell who they are, and the label doesn’t tell you what to change.

Checks to run before you use the word

1. Is the change bigger than the test’s noise? Know your test’s typical error, ideally from repeated measurements in your own setting, with your protocol and your kind of clients. The 3 to 5% figures above are illustrations, not a universal threshold. Treat a change below the bar as undetermined, not as zero. Clearing the bar doesn’t prove the training caused the change either, and falling short doesn’t prove the training did nothing. If the number is an estimate from a submaximal test, widen the margin further.

2. Did the dose actually happen? Count sessions completed, not sessions prescribed. Check the intensity reached, not the one written on the plan. In the Ross trial, everyone analyzed had completed at least 90% of sessions, and the low dose still left 38.5% without a measurable gain. Adherence matters, but a well-attended dose can still be too small.

3. Did you check the outcomes that matter to this client? Decide in advance which outcomes fit the client’s goals: submaximal heart rate at a fixed workload, a lactate or ventilatory threshold, a time trial, strength or a functional test. Judge each one against its own error. Checking ten outcomes until one looks better is its own trap, because noise alone will hand you a winner. What you’re after is a profile of what moved and what didn’t, not a verdict on the person.

4. Were the two tests comparable? ACSM’s Guidelines note that peak VO2 is typically 5 to 20% lower on a cycle ergometer than on a treadmill, and recommend using the same mode at each time point. A first test can also come out low because the person is still learning it, which is why some researchers treat it as a familiarization trial (Edgett et al., 2018). Sleep, caffeine, food and fluids, a recent illness, a new medication or a hard session the day before can each move a single result. So can the timing of the retest: too soon after the last hard week and fatigue masks the gain, too long after and detraining eats into it. And if you track VO2max in mL/kg/min, a change in body weight moves the number without any change in the heart or muscles.

Then make the decision the label never gives you. If the flat result comes with fatigue, pain, symptoms, a health change or a new medication, the next step is recovery or a referral, not more load. Otherwise, adjust the stimulus within what the client can tolerate: more sessions or more intensity, one change at a time. Or switch the mode: in a twin study, individual responses to endurance and resistance training weren’t correlated, so a low response to one mode didn’t predict a low response to the other. The authors suggest that low responders may be rescued by switching (Marsh et al., 2020). Give the program time, since responses in the older-adult resistance study were more positive after 24 weeks than after 12 (Churchward-Venne et al., 2015). Next time, plan two tests at baseline and two at follow-up, after a familiarization test, and compare averages. With two tests at each end, the error of the change drops to about one typical error, and the 95% bar falls from about 2.8 to about 2 typical errors. A single extra retest run only because the first result disappointed will tend to look better anyway, through regression to the mean.

Why this matters on the exam

If an ACSM-EP scenario describes a client who hasn’t progressed, read it as a request for a decision. Look for the option that checks what was delivered or how it was measured, or that adjusts the prescription. An answer that only labels the client doesn’t address the problem.

FAQ

What is an exercise non-responder? Someone whose change on a given outcome after a training program doesn’t clear a chosen threshold: zero in some studies, the test’s typical error in others. It’s usually judged on one outcome, such as VO2max, and one pre/post comparison. The term describes an observed result on one outcome in one program. It doesn’t show that the person can’t adapt, because a single comparison mixes the true response with measurement noise.

Do non-responders to exercise really exist? Differences in how much people gain from the same program probably exist, and they’re partly heritable. People who gain nothing on every outcome appear to be rare. In several trials every participant improved on at least one measure, and in others a higher training dose reduced the share classified as non-responders, in some groups to zero.

What percentage of people don’t respond to exercise? It depends on the outcome, the definition and above all the dose. In one study the rate ranged from 69% at one session a week to 0% at four or five sessions a week (Montero & Lundby, 2017). In another it went from 38.5% at a low dose to 0% at a higher amount and intensity (Ross et al., 2015). A figure quoted without the dose means very little.

How do you know if a client is responding to training? Compare each change with the typical error of the test, decide in advance which outcomes matter for the client, and check that the planned dose was delivered. Two tests at each time point, averaged, tell you more than a single before-and-after pair.

Can a non-responder become a responder? A person classified as a non-responder can later be classified as a responder. In one trial every non-responder was reclassified as a responder after six more weeks with two extra sessions a week. Without a control group, and with regression to the mean in play, that shows reclassification more than a proven rescue. In a twin study, a low response to one training mode didn’t predict a low response to the other. One study in older women found that most who hadn’t gained muscle mass still hadn’t after more volume, so it isn’t guaranteed.

Is VO2max trainability genetic? Partly. HERITAGE estimated the maximal heritability of the VO2max response at 47%. More recent controlled analyses suggest the true spread in trainability is smaller than raw pre/post data make it look, and the same person’s observed response to the same program is poorly reproducible.

Key takeaways

A flat retest doesn’t establish non-response. A laboratory VO2max test has a day-to-day error of roughly 3 to 6%, so a real gain of a few percent can vanish in one comparison.
To pass the usual 95% bar against test noise, a single pre/post change has to exceed about 2.8 times the test’s typical error, or about 2 times if you average two tests at each end.
The share classified as non-responders shrinks as the dose grows: from 44.9% to 19.3% with volume in one trial, from 38.5% to 0% with amount and intensity in another.
In several studies that measured multiple outcomes, every participant improved on at least one.
Some true differences in trainability probably exist, but they’re smaller than raw scatter suggests, and individual responses are poorly reproducible from one training block to the next.
Before using the label, check the noise, the dose delivered, the outcomes that matter and the testing conditions. Then adjust the program, safely.

References

American College of Sports Medicine. ACSM’s Guidelines for Exercise Testing and Prescription. 12th ed. Wolters Kluwer; 2025. Chapters 4 and 5.
Atkinson G, Batterham AM. True and false interindividual differences in the physiological response to an intervention. Exp Physiol. 2015;100(6):577-588. doi:10.1113/EP085070
Atkinson G, Williamson P, Batterham AM. Issues in the determination of “responders” and “non-responders” in physiological research. Exp Physiol. 2019;104(8):1215-1225. doi:10.1113/EP087712
Barber JL, Ruiz-Ramie JJ, Robbins JM, et al. Regular exercise and patterns of response across multiple cardiometabolic traits: the HERITAGE family study. Br J Sports Med. 2022;56(2):95-100. doi:10.1136/bjsports-2020-103323
Bonafiglia JT, et al. Inter-individual variability in the adaptive responses to endurance and sprint interval training: a randomized crossover study. PLoS One. 2016;11(12):e0167790. doi:10.1371/journal.pone.0167790
Bonafiglia JT, Swinton PA, Ross R, et al. Interindividual differences in trainability and moderators of cardiorespiratory fitness, waist circumference, and body mass responses: a large-scale individual participant data meta-analysis. Sports Med. 2022;52(12):2837-2851. doi:10.1007/s40279-022-01725-9
Bouchard C, An P, Rice T, et al. Familial aggregation of VO2max response to exercise training: results from the HERITAGE Family Study. J Appl Physiol. 1999;87(3):1003-1008. doi:10.1152/jappl.1999.87.3.1003
Churchward-Venne TA, Tieland M, Verdijk LB, et al. There are no nonresponders to resistance-type exercise training in older men and women. J Am Med Dir Assoc. 2015;16(5):400-411. doi:10.1016/j.jamda.2015.01.071
Clarke J, et al. Comparison of measures of maximal and submaximal fitness in response to exercise. Med Sci Sports Exerc. 2017. doi:10.1249/MSS.0000000000001164
Del Giudice M, Bonafiglia JT, Islam H, et al. Investigating the reproducibility of maximal oxygen uptake responses to high-intensity interval training. J Sci Med Sport. 2020;23(1):94-99. doi:10.1016/j.jsams.2019.09.007
Edgett BA, et al. Reproducibility of peak oxygen consumption and the impact of test variability on classification of individual training responses in young recreationally active adults. Clin Physiol Funct Imaging. 2018;38(4):630-638. doi:10.1111/cpf.12459
Katch VL, Sady SS, Freedson P. Biological variability in maximum aerobic power. Med Sci Sports Exerc. 1982;14(1):21-25. doi:10.1249/00005768-198201000-00004
Marsh CE, et al. Fitness and strength responses to distinct exercise modes in twins: Studies of Twin Responses to Understand Exercise as a THerapy (STRUETH) study. J Physiol. 2020. doi:10.1113/JP280048
Metcalfe RS, Vollaard NBJ. Heterogeneity and incidence of non-response for changes in cardiorespiratory fitness following time-efficient sprint interval exercise training. Appl Physiol Nutr Metab. 2021;46:735-742. doi:10.1139/apnm-2020-0855
Meyler SJR, Swinton PA, Bottoms L, et al. Changes in cardiorespiratory fitness following exercise training prescribed relative to traditional intensity anchors and physiological thresholds: a systematic review with meta-analysis of individual participant data. Sports Med. 2025;55(2):301-323. doi:10.1007/s40279-024-02125-x
Montero D, Lundby C. Refuting the myth of non-response to exercise training: “non-responders” do respond to higher dose of training. J Physiol. 2017;595(11):3377-3387. doi:10.1113/JP273480
Nunes JP, et al. Responsiveness to muscle mass gain following 12 and 24 weeks of resistance training in older women. Aging Clin Exp Res. Published online 2020. doi:10.1007/s40520-020-01587-z
Odden I, et al. Limited reproducibility of individual physiological adaptations to repeated endurance exercise training. J Appl Physiol. 2026. doi:10.1152/japplphysiol.00154.2026
Pickering C, Kiely J. Do non-responders to exercise exist, and if so, what should we do about them? Sports Med. 2019;49(1):1-7. doi:10.1007/s40279-018-01041-1
Renwick JRM, Preobrazenski N, Wu Z, et al. Standard deviation of individual response for VO2max following exercise interventions: a systematic review and meta-analysis. Sports Med. 2024;54(12):3069-3080. doi:10.1007/s40279-024-02089-y
Reuter M, Rosenberger F, Barz A, et al. Does higher intensity increase the rate of responders to endurance training when total energy expenditure remains constant? A randomized controlled trial. Sports Med Open. 2023;9:35. doi:10.1186/s40798-023-00579-3
Ross R, de Lannoy L, Stotz PJ. Separate effects of intensity and amount of exercise on interindividual cardiorespiratory fitness response. Mayo Clin Proc. 2015;90(11):1506-1514. doi:10.1016/j.mayocp.2015.07.024
Ross R, Goodpaster BH, Koch LG, et al. Precision exercise medicine: understanding exercise response variability. Br J Sports Med. 2019;53(18):1141-1153. doi:10.1136/bjsports-2018-100328
Scharhag-Rosenberger F, Walitzek S, Kindermann W, et al. Differences in adaptations to 1 year of aerobic endurance training: individual patterns of nonresponse. Scand J Med Sci Sports. 2012;22(1):113-118. doi:10.1111/j.1600-0838.2010.01139.x
Shephard RJ, Rankinen T, Bouchard C. Test-retest errors and the apparent heterogeneity of training response. Eur J Appl Physiol. 2004;91:199-203. doi:10.1007/s00421-003-0990-4
Sisson SB, Katzmarzyk PT, Earnest CP, et al. Volume of exercise and fitness nonresponse in sedentary, postmenopausal women. Med Sci Sports Exerc. 2009;41(3):539-545. doi:10.1249/MSS.0b013e3181896c4e
Zinner C, et al. The magnitude and time-course of physiological responses to 9 weeks of incremental ramp testing. Scand J Med Sci Sports. 2023. doi:10.1111/sms.14347

Preparing for the ACSM-EP?

Engram Kinetics trains the decisions the exam asks for across its four domains: 41 guided lessons, 56 Engrams, 11 masterclasses and 5 integrated case studies. Every wrong option is tied to a named reasoning error, and the feedback walks through the reasoning, not only the key. $199, one payment. Lifetime access to the program you bought, including its future updates.

Start the ACSM-EP preview See the ACSM-EP program

Free account needed for the preview.