Cardiorespiratory Fitness and Mortality established that CRF predicts mortality — but, being cross-sectional and observational, it could not say CRF is a lever rather than a marker. This AHA scientific statement (Ross 2016) (Ross et al., 2016) supplies the three things that make CRF actionable: you can measure it cheaply, you can raise it a known amount with a known exercise dose, and raising it tracks lower risk. Together they move CRF from marker toward modifiable target — though, as below, still short of RCT-proven causality on hard outcomes.
Evidence-tier note (confidence: medium, raised from low 2026-08-06). The raise-it leg is now
gold-backed: Poon 2024, an umbrella of 24 SRs+MAs (429 primary studies, 12 967 participants), confirms
the exercise→CRF dose from a different body than the AHA and adds the HIIT-vs-MICT head-to-head. The
measure-it and vital-sign legs still rest on the single AHA scientific statement (Ross 2016 —
consensus tier), and the causal leg (raising CRF → lower mortality) stays overwhelmingly
observational. So two of three legs are well-supported and one (causality) is not — medium, not high.
(Poon et al., 2024; inferred from Ross et al., 2016)
The measure-it leg now also carries one measured-VO2 cohort for the cycle route (Kokkinos 2018, FRIEND
registry), but it shares three authors with Ross, so it adds data, not independent backing; the grade
stays medium. (Kokkinos et al., 2018; inferred from Ross et al., 2016)
A walk-test route now rests on Oja 1991 (UKK 2-km walk test derivation and cross-validation), a different
group with no shared authors, but a single 1991 validation study (64 people in the derivation); it adds
a route, not a grade change. (inferred from Oja et al., 1991)
Marín-Jiménez 2023 (n=410, Spain) adds out-of-sample agreement and test-retest data for Oja’s equation;
it tests that equation rather than backing it independently, so it bounds the walk route’s per-person
error without changing the grade. (inferred from Marín-Jiménez et al., 2023)
Measure it — three tiers, and a cheap one that works
- CPX (cardiopulmonary exercise test) — direct peak VO2, the gold-standard, most accurate and standardized quantification. Needs equipment and trained staff.
- Maximal exercise test without gas analysis — CRF estimated from peak treadmill/cycle work rate. Note the modality effect: cycle-ergometer values run «10% to 20% lower when using a cycle ergometer compared with a treadmill in untrained individuals».
- Non-exercise estimated CRF (eCRF) — the cheap route: 13 cross-validated equations predict CRF from «readily available clinical variables» (the Jurca/Nes inputs are typically age, sex, BMI, resting heart rate and self-reported activity), no exercise test needed. The Jurca (2005) and Nes (2011) models predict long-term mortality comparably to measured CRF: per-1-MET risk reduction «7.4% to 21%» (all-cause) and «8% to 16.9%» (CVD) — bracketing the measured-CRF meta-analytic 13%/15% (Cardiorespiratory Fitness and Mortality, Kodama). Hard caveat: eCRF «should not be viewed as a replacement for objective assessment of CRF» in at-risk patients.
Estimating VO2max from a maximal cycle test — the FRIEND equation vs the ACSM equation
The equation and where it comes from. Kokkinos 2018 fitted a cycle equation to directly measured VO2max in the FRIEND registry: 5,100 adults (3,378 men, mean age 35.9; 1,722 women, mean age 47.5; range 18-87; 4,617 White; 57% overweight or obese), each on a graded maximal cycle test in one of eight US/European labs, with maximal effort required («a peak respiratory exchange ratio ≥1.0») and people with CVD, cancer, COPD, CKD or PAD excluded (but «subjects with particular conditions (e.g., diabetes and obesity), musculoskeletal concerns (e.g., back pain and osteoarthritis), and CVD risk factors» kept in). Result: «VO2 max in ml O2●kg-1●min-1= 1.74* [Watts*6.12/body weight (kg)] +3.5», with sex-specific slopes of 1.76 (men) and 1.65 (women). (Kokkinos et al., 2018)
The ACSM equation over-reads by about 15% on average. Bias was measured as (predicted - measured) / measured. «The ACSM equation overestimated VO2 max by over 15% for the entire cohort (average relative bias: 15.46 % ± 0.13)» — +11.23% in men, +23.77% in women — against +0.51% for the FRIEND equation. The single FRIEND equation still splits by sex (-0.78% men, +3.04% women); the sex-specific versions give +0.26% (men) and -1.37% (women). The 30% hold-out sample repeats the pattern (+15.78% vs +0.59%). All of these are cohort means inside FRIEND. The authors’ explanation: the ACSM equations come from submaximal steady-state work in mostly young subjects, extrapolated to a maximum where «a steady-state is rarely achieved at high levels of exercise». (Kokkinos et al., 2018)
What that means for a bike reading. Since 1.74 x 6.12 ≈ 10.65, the equation is VO2max ≈ 10.65 x (peak watts / kg) + 3.5 ml·kg-1·min-1, or about 3.04 x W/kg + 1 in METs. A 250 W peak at 80 kg (3.1 W/kg) gives about 36.8 ml·kg-1·min-1, about 10.5 METs. A readout built on the ACSM equation would put the same ride higher. In the cohort the ACSM equation read above measured VO2max by about 11% in men and 24% in women on average, and above the single FRIEND equation’s mean estimate by about 12% (men) and 18% (women) (Table 3A means: 46.77 vs 41.94; 28.22 vs 24.01). (inferred from Kokkinos et al., 2018)
What the paper does not give.
- No usable individual error. Table 3 is headed «Mean ± Standard Deviation», and each mean relative
bias carries an SD (FRIEND 0.11, ACSM 0.13), which would describe per-person spread. Its scale is
unclear: the text writes «0.51% ± 0.11%», and an SD of 0.11 percentage points is
implausibly small for single tests; read as a fraction it would mean roughly 11-13% spread for one
person, close to Ross’s submaximal SEE. No standard error of the estimate or limits of agreement is
given [searched: standard error of the estimate / limits of agreement / Bland / correlation across
chunk 01, 0 hits; positive control
relative bias13 hits]. So a near-zero average bias does not say how far one person’s estimate may sit from their measured value. (Kokkinos et al., 2018) - Maximal tests only. The input is the work rate of a graded test to exhaustion. Watts from a submaximal steady session (say 15 min at 75% HRmax) fall outside the data the equation was fitted to; that case is the submaximal HR-extrapolation route above, with its own error. (inferred from Kokkinos et al., 2018)
- Protocol not specified. Protocols were lab-specific, and the paper names no ramp rate or stage length [searched: ramp / increment / stage / peak work across chunk 01, 0 hits]. Peak watts on a step protocol depend on stage length, so a home ramp test may not match the labs’ inputs. (inferred from Kokkinos et al., 2018)
- Validation inside one registry only. The authors write: «Additional studies are needed in different populations to further explore the portability of the FRIEND equation.» (Kokkinos et al., 2018). The held text is the accepted manuscript, which contains internal slips (the hold-out is headed n=1,530, 30% of 5,100, but its “All” row reads n=1,758; the Conclusion labels the single equation’s -0.78% / +3.04% as the sex-specific results). The women in the cohort were older and less fit than the men, so the larger ACSM error in women cannot be pinned on sex. (inferred from Kokkinos et al., 2018)
Matched against Ross’s error figures (NO CELL, NO CLAIM).
| Parameter | Ross 2016 | Kokkinos 2018 | Same quantity? |
|---|---|---|---|
| Test | submaximal: 2 steady-state work rates, HR line extrapolated to age-predicted HRmax | graded maximal cycle test, measured VO2max | NO |
| Error figure | SEE «±10% to 15%» (spread for one person) | mean relative bias +0.51% FRIEND, +15.46% ACSM (average offset); SD 0.11 / 0.13, scale unclear | NO — spread vs offset; the SD’s scale cannot be matched |
| Cycle comparison | cycle 10-20% below treadmill, untrained (modality gap) | cycle equation vs cycle measurement (equation gap) | NO |
| Authors | Ross, Blair, Arena, Kaminsky, Myers, et al. | Kokkinos, Kaminsky, Arena, Zhang, Myers | shared authors — not independent |
No row matches, so the two figures do not combine into one error band. Ross bounds the per-person error of the submaximal route. Inside the FRIEND cohort, Kokkinos removes the ACSM equation’s average +15% offset from the maximal route (the single equation still reads +3% in women), but its per-person error cannot be read from the paper. The individual error of a maximal cycle test read through the FRIEND equation is an open gap (type-G): no held source reports it in usable form. — an agreement study (SEE or limits of agreement) of FRIEND-estimated vs measured cycle VO2max. (Kokkinos et al., 2018; inferred from Ross et al., 2016)
Placing a bike number on Mandsager’s bands. Mandsager’s METs were «determined based on treadmill grade and speed at peak exercise», and the paper names no conversion equation (Mandsager et al., 2018). Two offsets point the same way: cycle values run below treadmill values in the untrained (Ross), and if Mandsager’s conversion behaves like the steady-state equations Kokkinos criticises, its METs run high. Kokkinos’s own data are cycle. It prints its group’s FRIEND treadmill equation and says the ACSM treadmill error is «>4 times» larger, without giving the sign; the one direction it states, about the ACSM equations generally, is age-bound: «VO2 max will likely be overestimated by the ACSM equations when applied to older populations» (Kokkinos et al., 2018). A FRIEND-estimated or measured cycle VO2max placed on that grid would then understate a person’s band, by an amount no held source gives. What would settle it: one cohort with both measured cycle VO2max and treadmill-estimated METs in the same people. (Kokkinos et al., 2018; Mandsager et al., 2018; inferred from Ross et al., 2016)
Where your number sits: the FRIEND registry «published peak VO2 reference standards for adult men and women (20–79 years of age)» — the US normative percentiles (the sibling of Mandsager’s percentile MET grid). About «half of the variance in CRF is considered to be attributable to heritable factors», so a fraction of your position is not trainable.
Estimating VO2max from a walk — the UKK 2-km walk test
Where walk tests sit. Ross lists field tests beside the lab routes. Running tests need near-maximal effort; «A modification designed to limit the exercise inten- sity, and thus make it more widely applicable, is the 1-mile walk test.» Their advantage is that «they require minimal resources (measured course, timing device, and palpated pulse rate) and can be self-administered». For patients who are «markedly deconditioned» the 6-minute walk test is common, but it «may not necessarily provide an ac- curate estimation of CRF»; others found it «provided a reasonable estimate of CRF» in outpatients with stable coronary heart disease, so the evidence on it is mixed. (Ross et al., 2016)
The test and the equations. Oja 1991 (UKK Institute, Tampere) compared 1, 1.5 and 2 km walks and kept 2 km: most people preferred it, and walk time tracked VO2max at least as well as the shorter distances. The instruction was «Walk the distance as fast as you can, but do not risk your health»; heart rate is the mean over the last 30 s. Sex-specific equations, VO2max in ml·kg-1·min-1 (Table 7, BMI model; read from the page image of the 1991 scan):
- men: 184.9 − 4.65 × time − 0.22 × HR − 0.26 × age − 1.05 × BMI (r² 0.75, SEE 5.1, n = 34)
- women: 116.2 − 2.98 × time − 0.11 × HR − 0.14 × age − 0.39 × BMI (r² 0.73, SEE 3.3, n = 28)
The paper puts the SEE at «12—14% for men and 9% for women» of the mean; the abstract gives «of the order of 9—15% of the mean». These are in-sample figures from the derivation group. Heart rate alone «correlated only weakly with VO2max»; it adds precision in combination with time and age (Table 5). A body-weight version exists too (Table 6; r² 0.66 men, 0.76 women). (Oja et al., 1991) The time unit is printed «msec». Entering the men’s group means with time in decimal minutes (15.2 min, HR 153, age 43, BMI 25.1) gives about 43.0 against 43.1 measured, so minutes is the working reading. A regression evaluated at its own means returns about the mean by construction, so the check’s force is only that seconds or milliseconds would give impossible values. (inferred from Oja et al., 1991)
Derived in whom. 64 working adults from Tampere, Finland (35 men, 29 women), age 20-65 in four bands, with only 3 women and 7 men aged 60-65. All were medically screened and «considered fit for brisk walking». Their VO2max was measured on a treadmill walk to maximal effort: 34.8 ± 6.7 ml·kg-1·min-1 in women, 43.1 ± 9.9 in men. By heart rate the walk is vigorous (the wiki’s label; finishing heart rate averaged about 85% in women and 87% in men of measured HRmax), though on average subjects rated the effort moderate on Borg’s 0-10 scale. The authors call the medical screening an «inevitably selected study group». (Oja et al., 1991)
In new groups it read low. Five separate groups of healthy adults walked twice and had VO2max measured (Table 8, BMI model). «The estimated mean VO2max was sys- tematically smaller than the measured value in all cross-valida- tion groups», by about 3% (obese women, BMI 27-40) to 12% (obese men) of measured ml·kg-1·min-1. The measured-vs-estimated correlation was 0.75-0.79 in obese women, obese men and moderately active men, 0.55 in moderately active women, and 0.53 in highly active men (measured 57.6 ml·kg-1·min-1). The authors’ verdict: valid for «healthy obese and nonobese adult men and women but not that of very active and fit men». The paper prints no SEE for the cross-validation groups [searched: standard error / Bland / limits of agreement across chunk 01; Table 8 columns checked on the page image]. (Oja et al., 1991) (The % offsets are arithmetic on the printed group means,.)
Do one practice walk. On a repeat walk people went about 9-10 s faster and predicted VO2max rose 1.6-2.2%, a small but significant shift. «This ob- servation implies that for accurate V02 estimation the first test should be conducted for familiarization purposes.» Test-retest r was 0.98 (men) and 0.94 (women). (Oja et al., 1991)
One person’s walk estimate: about ±12.6 ml·kg-1·min-1 either way (Marín-Jiménez 2023). A Cadiz group tested Oja’s equation, unchanged, in 410 Spanish adults aged 18-64, balanced by sex, three age bands and self-reported activity (active = meeting WHO guidance), against treadmill VO2max measured by gas analysis (mean 36.1 ml·kg-1·min-1). On average the estimate was close: «The difference between measured and estimated VO2max by Oja’s equation was −0.30 ml* kg−1 * min−1 (95% LoA = −12.86 to 12.26, p < 0.001)»; r 0.78; SEE 5.0 ml·kg-1·min-1 for the line refitted to this sample (measured = 12.24 + 0.66 × estimate). Finishing heart rate came from a Xiaomi Mi Band 4 wrist band, which the authors describe as «validated in adults», and no practice walk is described. (Marín-Jiménez et al., 2023) So for 95% of people the estimate sat within about 12.6 ml·kg-1·min-1 above or below their measured value — about ±35% of the sample mean, or roughly ±3.6 METs. Applied unchanged, the equation’s per-person SD of the difference is about 6.4 ml·kg-1·min-1 (LoA half-width / 1.96), about 18% of the mean, sexes pooled: above Oja’s in-sample 9-14%. The printed SEE of 5.0 (14%) holds only after refitting the line to this sample. The authors call these limits «narrow» and the test valid; ±35% of the mean is the wiki’s reason to read one walk as a band, not a number. (inferred from Marín-Jiménez et al., 2023)
Worse agreement in the fit. The gap grew with VO2max (heteroscedasticity: difference vs mean of measured and estimated, R² 0.117). In active participants the mean difference was −3.36 (95% LoA −15.31 to 8.59), in non-active −0.19 (−13.45 to 13.06); the authors read it as walking not stressing a fit person enough «(i.e., underprediction)». (Marín-Jiménez et al., 2023) The direction is not certain: neither text nor plot says which value is subtracted, and Table 1’s means put the estimate slightly above measured in both groups (active 40.14 vs 39.67). The scatter line in Fig 1A, measured = 12.24 + 0.66 × estimate, says that at group level a high walk estimate tends to sit above measured and a low one below: an estimate of 50 matches about 45 measured on average, an estimate of 25 about 29. By activity group the active band was not wider (width 23.9 vs 26.5) but shifted; the worse-in-the-fit reading rests on the heteroscedasticity regression. (inferred from Marín-Jiménez et al., 2023)
Oja’s average under-read did not recur. Oja’s five groups read 3-12% low on average; here the group-mean estimate was about 1% above measured (36.58 vs 36.13). The two studies differ in criterion protocol, country and heart-rate device, so which of these explains the difference cannot be told from the two papers. (Marín-Jiménez et al., 2023; inferred from Oja et al., 1991)
Tracking change: two walk estimates are noisy. Retested about a week later, walk time barely moved (T2-T1 −1.48 s, 95% LoA −21.07 to 18.10 s on a mean of about 998 s; ICC 0.99), and the mean estimate barely moved (−0.29; ICC 0.95; Table 3 prints p = 0.206, the Bland-Altman text p = 0.034 for the same difference). But the estimate’s 95% LoA was −8.13 to 7.54 ml·kg-1·min-1. The paper gives no minimal detectable change [searched: minimal detectable / smallest / MDC, 0 hits; positive control Bland 14 hits]. (Marín-Jiménez et al., 2023) Read as a noise band, in 95% of people two estimates a week apart differed by less than about 8 ml·kg-1·min-1 (about ±22%). Ross’s «≈10% improvement in CRF in previously sedentary adults» from meeting activity guidance (see Raise it) is about 3.2 ml·kg-1·min-1 at this study’s non-active mean (32.2), well inside that band, so one before-and-after pair of walks cannot confirm a typical training gain. By Oja’s coefficients a 20-s time difference moves the estimate only about 1-1.8 ml·kg-1·min-1 (BMI or weight model), so, if the paper applied the equation as printed, most of the retest spread would have to come from finishing heart rate; the paper reports no heart-rate retest figure. The retest data cover a week’s noise, not whether a change in the estimate follows a change in measured VO2max. (inferred from Marín-Jiménez et al., 2023; Oja et al., 1991; Ross et al., 2016) The authors suggest pacing may explain the walk’s weaker reliability, a cause that would itself act partly through finishing heart rate: «starting too fast, so that the participants are unable to maintain their speed throughout the test; or too slow, increasing their speed at the end of the test (which may also mean an unexpected incremented heart rate at the end of the test).» Device noise and uneven pacing were not tested separately. (Marín-Jiménez et al., 2023)
No better walk equation; the shuttle run reads tighter. Adding sex, age, activity level and %BF to time and heart rate raised in-sample R² to 0.70 (SEE 4.4), but the authors concluded «our results did not improve the prediction equation proposed by Oja et al.» and print no usable new equation. In the same people the 20-m shuttle run (Léger equation, not printed in the paper) agreed more closely (95% LoA −7.30 to 9.02) and retested more tightly (−1.51 to 1.58); the authors favour it, with the walk «more suitable for adults who are unable to run». (Marín-Jiménez et al., 2023)
Oja and Marín-Jiménez matched (NO CELL, NO CLAIM).
| Parameter | Oja 1991 | Marín-Jiménez 2023 | Same quantity? |
|---|---|---|---|
| Equation | derived (sex-specific, BMI and weight models) | Oja’s, model variant not stated | YES, variant |
| Criterion | treadmill walk to max, gas analysis | treadmill CPET (walk or run protocol), gas analysis | similar, protocols differ |
| Heart-rate device | Polar telemetric cardiometer (Sport Tester PE 3000) | Xiaomi Mi Band 4 wrist band | NO |
| Per-person spread | SEE 3.3 / 5.1 (9%, 12-14%), fitted in the derivation sample, by sex | SD of difference ≈6.4 (≈18%), equation unchanged; SEE 5.0 (14%) after refitting; sexes pooled | SEE vs SEE: same kind (fitted residual), unmatched samples; SD of difference vs derivation SEE: comparable as prediction error (out-of-sample vs fitted; mean bias removed in both) |
| Group-mean offset, new people | −3% to −12%, five groups | about +1% (Table 1 means) | YES (ratio of group means) |
| Retest shift | +1.6-2.2% estimate, 9-10 s faster (4th vs 5th walk) | −0.29 estimate, −1.48 s (T1 vs T2; no earlier walk described) | mean T2-T1, but NOT the same walk history |
| Retest agreement | r 0.94-0.98 | ICC 0.95; LoA −8.13 to 7.54 | NO (r vs ICC) |
| Authors | Oja, Laukkanen, Pasanen, Tyry, Vuori | Marín-Jiménez, Castro-Piñero, Cuenca-García, et al. | no shared author |
Marín-Jiménez tests Oja’s own equation, so its agreement with Oja is a refinement (it adds the out-of-sample per-person error and retest band Oja lacked), not independent corroboration. Matched on the rows that line up: the per-person error of the unchanged equation in a new population is larger than Oja’s in-sample figures, and the average under-read did not recur. The retest shifts are not comparable: Oja’s people had already walked several times; Marín-Jiménez describes no earlier walk, and still shifted only 1.5 s. (Marín-Jiménez et al., 2023; inferred from Oja et al., 1991)
Matched against the bike routes (NO CELL, NO CLAIM).
| Parameter | Ross 2016 | Kokkinos 2018 | Oja 1991 | Same quantity? |
|---|---|---|---|---|
| Test | submaximal, HR extrapolated | maximal cycle test | fast self-paced 2-km walk | NO |
| Spread for one person | SEE «±10% to 15%», a general range | none usable (SD scale unclear) | SEE 9% (women), 12-14% (men), in-sample | Ross vs Oja: same kind (SEE as % of mean), not matched samples; Kokkinos: no cell |
| Average offset, new people | — | +0.59% FRIEND, +15.78% ACSM (30% hold-out) | −3% to −12%, five external groups | NO — mean of per-person ratios vs ratio of group means; internal hold-out vs external groups |
| Authors | Ross, Blair, Arena, Kaminsky, Myers, et al. | Kokkinos, Kaminsky, Arena, Zhang, Myers | Oja, Laukkanen, Pasanen, Tyry, Vuori | Oja shares no author with either |
Only the spread row lines up. The walk test’s in-sample SEE (9% women, 12-14% men) is about the ±10-15% Ross gives for submaximal tests, so on spread a fast walk is the same class of estimate as a submaximal heart-rate test, not a sharper one. Its out-of-sample SEE is not reported, and Oja’s own data suggest new-group error is larger: the measured-vs-estimated correlation fell from about 0.86 in the derivation group (r² 0.73-0.75) to 0.53-0.79 in the five new groups. This is not type-E corroboration: Oja makes the same comparison itself (its results «compared favourably» with submaximal tests), and Ross’s figure is a general range, not an independent measurement of this test. The offsets cannot be pooled across the two papers. Within Oja, the group-mean estimate ran low in all five groups, by up to about 12% in obese men; the paper gives no per-person direction or spread. (Kokkinos et al., 2018; Oja et al., 1991; inferred from Ross et al., 2016)
Who it suits, on this evidence. Healthy adults who can walk 2 km briskly and have no lab or bike. The external check covered ages 20-65 only in the obese groups (BMI 27-40); the non-obese groups were 35-45, and the derivation had few over-60s. Obese men, though called «reasonable» by the authors, carried the largest average under-read (12%). Outside the data: over-65s, people with heart or lung disease, people on heart-rate-modulating drugs (medication was an exclusion at the treadmill stage), and very fit men, where the paper says it fails. (inferred from Oja et al., 1991) Marín-Jiménez extends the tested range to 18-64 in a second country, both sexes and non-active as well as active adults; over-65s remain untested in both papers. On Mandsager’s treadmill-based MET bands, the average offset of a walk estimate is not settled (low in Oja’s groups, near zero in Marín-Jiménez’s), and per person the 95% band is about ±3.6 METs, so a single walk can place someone in the wrong band. (Marín-Jiménez et al., 2023; inferred from Oja et al., 1991)
Raise it — the exercise dose → CRF response
Meeting consensus physical-activity recommendations buys «≈10% improvement in CRF in previously sedentary adults». Both amount and intensity move it, but not equally: «Increases in CRF appear more responsive to increases in intensity than increases in session duration or frequency».
The actionable rule is baseline-stratified intensity — the fitter you already are, the harder you must work to gain:
| Baseline CRF | Intensity needed for a clinically meaningful (≥1 MET) gain |
|---|---|
| < 10 METs | ≈50% of heart-rate reserve / VO2 reserve is adequate |
| 10–14 METs | 65–85% of HR reserve / VO2R |
| > 14 METs | > 85% — and at >=13 METs (Ross’s own, slightly lower cutoff for this separate point) the goal is «more related to improving performance than health» |
Worked magnitudes: at fixed 50% intensity — 50% of peak VO2, a different scale from the HRR / VO2R reserve percentages in the table above — 30 min×5/wk gave a 9.4% CRF rise vs 15.6% for 60 min; raising intensity to 75% of peak VO2 gave 19.6% (Ross 2015, as reported in Ross’s narrative). The same trial’s row in Ross’s Table 7 lists 7.7% (LALI) and 14.8% (HALI) for the two 50% arms, with 19.6% (HAHI) matching; the source does not reconcile the narrative and table figures, so both are kept here. (corrected 2026-10-06: added the peak-VO2 scale and the Table 7 7.7%/14.8% figures beside the narrative 9.4%/15.6%; Ross chunk 02) STRRIDE showed a clean dose gradient — 6% (low amount / moderate intensity), 11% (low / high), 18% (high / high). Interval training beats equal-energy continuous training: in one trial near-maximal intervals gave «20.6%» vs «9.4%» for moderate continuous. Older adults gain too (a 41-trial meta-analysis: +16.3%).
Holding the effort target as you get fitter
What the target is measured against. Heart-rate reserve (HRR) is maximum heart rate minus resting heart rate; a target of x% of HRR is resting HR plus x% of that span (the Karvonen form). Ross uses the term (abbreviation list: «HRR, heart rate reserve») but does not define it in the text read. (inferred from Ross et al., 2016) [searched: heart rate reserve / HR reserve across Ross chunks 01-03]
Training raises fitness without raising maximum heart rate. Ross: «Because virtually every exercise training study, regardless of length or intensity, has reported no change or even a slight decline in HR max, increases in CRF occur primarily via increases in stroke volume, arteriovenous O2 difference, or both.» (Ross et al., 2016)
- Consequence — a relative target self-progresses. If HRmax stays put while stroke volume and O2 extraction rise, the same heart rate is reached only at a higher absolute work rate (more watts, faster pace) as fitness improves. A target fixed as a % of one’s own HRR therefore asks for more absolute work over time, raising the workload without a separate progression rule. (inferred from Ross et al., 2016)
- What would confirm or refute it: a trial logging work rate at a fixed %HRR across a training block (absolute work at the target HR should rise in step with measured VO2peak), or showing that a fall in resting HR or HR drift erodes the absolute load the % target implies. No held source reports this. (inferred from Ross et al., 2016)
Why the target is stratified by baseline, and when a step-up is due. Ross stratified its dose review because of «the significant role baseline CRF plays in the absolute intensity of the exercise regimen» (Ross et al., 2016) — the review bands were «(1) low (<9 METs); (2) intermediate (9–14 METs); and (3) high (≥15 METs)», which differ slightly from the recommendation bands in the table above (<10 / 10-14 / >14 METs, «The higher the baseline CRF, the more vigorous the intensity needed to produce a clinically significant increase in CRF»). (Ross et al., 2016)
- Where the relative target must be raised deliberately: the self-progression above holds within a band. The point where a person needs to raise the relative effort (from ≈50% HRR toward 65-85%) is when they leave the <10 MET band and still want further CRF gains. (inferred from Ross et al., 2016)
- On the mortality evidence that step-up may be optional. Ross: «Most of the lower mortality risk associated with a higher CRF occurs by the time a CRF of 10 to 12 METs is achieved. CRF values >12 METs are associated with a relatively lower impact on risk of all-cause and CVD mortality.» (Ross et al., 2016) Read with Ross’s statement, staying at ≈50% HRR once past ~10 METs is arguable on mortality grounds rather than a failure, with a smaller mortality return from higher bands, though Mandsager saw a further elite-vs-high step in patients 70+ (HR 0.71, 0.52-0.98) (Mandsager et al., 2018); its bands are age/sex-relative and elite starts at 8.4-11.5 METs at 70+ (Mandsager et al., 2018), so that step is not cleanly above 10 METs. The basis is a statement-level sentence with no citation of its own; the only citation in the three sentences ending at ref 290 (on the closing sentence) is Arem 2015, titled as a pooled dose-response analysis of leisure time physical activity and mortality (Ross et al., 2016); by its title an activity study rather than a fitness study (corrected 2026-10-22). Whether ≈50% HRR holds CRF at that level, and what the step-up does for daily function, no held source reports (see also Physical Activity Dose and Mortality). (inferred from Ross et al., 2016)
Heart-rate targets on heart-rate-modulating drugs. Of the submaximal test that estimates CRF from the work-rate/HR relation, Ross states: «This method cannot be applied with patients using HR-modulating medications (eg, β-blockers).» (Ross et al., 2016) — said of CRF estimation, not of training prescription.
- Extension to training targets: by the same mechanism (the drug blunts the HR response to work), a %HRR training target is equally unreliable for someone on a β-blocker, and effort is then gauged by perceived exertion (e.g. a rating-of-perceived-exertion scale) instead. (inferred from Ross et al., 2016)
Gaps (G). Two questions bear directly on how the target is set and held; the held fabric answers neither. This is a gap in what the wiki holds, not a finding about the literature: no literature search was run for either question, so it is not known whether head-to-head trials exist, and the absence here is not a null result. No direction is asserted.
- (a) Measured vs age-estimated HRmax. No held source shows whether anchoring %HRR on a measured (exercise-test) HRmax rather than an age-predicted one changes training outcomes at moderate intensity. Ross uses «age-predicted maximal HR» only inside the submaximal CRF-estimation method (chunk 01). (inferred from Ross et al., 2016)
- (b) Progressing vs fixed relative intensity. No held trial compares aerobic training that raises relative intensity over time against training held at a fixed %HRR, nor shows whether CRF plateaus on a fixed relative dose. Held trials include progressive protocols — HERITAGE trained at «55% to 75% of maximal HR» with «session duration and intensity progressively increased approximately every 2 weeks» (Ross et al., 2016) — but as a single arm, not a contrast.
HIIT vs continuous training — the gold umbrella (Poon 2024)
Ross’s single-trial interval signal (20.6% vs 9.4% above) is now backed by an umbrella review — «all reviews consistently demonstrated that HIIT significantly improves CRF» vs non-exercise control («SMD… 0.28 to 4.31… WMD… 3.25 to 5.5 mL/kg/min»), and head-to-head «the majority of reviews indicated that HIIT leads to similar or greater improvements in CRF» than moderate-intensity continuous training (MICT) («SMD… 0.18 to 0.99… WMD… 0.52 to 3.76 mL/kg/min»). (Poon et al., 2024)
Read the head-to-head edge as real but modest, and smallest where fitness is already normal. The HIIT-over-MICT SMD is 0.04–0.64 in healthy adults and 0.26–0.99 in overweight/obesity, and sprint interval training vs MICT is essentially a wash (SMD 0.04–0.18) — so the practical case for HIIT is time-efficiency (comparable or slightly greater CRF for less total exercise time), not a large fitness advantage. Safety is not a differentiator: «The safety concerns associated with HIIT do not appear to be significantly greater than those associated with MICT» (compliance generally ≥80%; pre-screen inactive at-risk individuals). (Poon et al., 2024)
This refines, it does not contradict, the single-trial number (parameter-table check): Ross’s «20.6% vs 9.4%» is a within-arm percent change in one trial; Poon’s «0.18 to 0.99» is a between-group standardized difference pooled across reviews — different quantities, so the umbrella upgrades the evidence grade and bounds the effect (modest between-modality gap), rather than clashing with it. Certainty caveat: «Most of the systematic reviews received moderate-to-critically low AMSTAR-2 scores», so the direction is robust but the constituent reviews are low-certainty. Poon is an umbrella of HIIT→CRF intervention reviews — a different question from the held CRF→mortality cohorts (Kodama / Mandsager), so it is a type-F upgrade of the raise-it leg, not independent backing for the causal leg below. (Poon et al., 2024)
One coronary-rehab study, read two ways: absolute vs relative risk [2026-10-21, Franklin]. Poon’s
safety sentence sits beside its own caveat and extends to high-risk groups: «while large-scale studies with
safety outcomes are currently lacking, some clinical studies have suggested that HIIT appears to be safe,
well-tolerated, and attainable, even when applied to relatively high-risk pop- ulations with low initial
fitness.» Its citations for that (refs 53-57) include Rognmo 2012, the coronary-rehab event-rate study, and
its small-trial example of four adverse events across 14 RCTs and 429 middle-aged and older adults
includes a myocardial infarction. (Poon et al., 2024)
(Poon et al., 2024)
An AHA scientific statement (consensus) cites the same study (n=4846 coronary patients in cardiac rehab):
«The absolute risk was low for HIIT (1 event per 23 182 patient-hours) and MICT (1 event per 129 456
patient-hours) but 5.6 times higher in the HIIT group.» It gives no event counts or confidence interval
for the ratio, and it also says «Data on the safety of HIIT in clinical populations are scarce.»
(Franklin et al., 2020)
| Parameter | Poon 2024 (umbrella) | Franklin 2020 (AHA statement) | Same quantity? |
|---|---|---|---|
| Population | includes «relatively high-risk» groups | coronary patients | overlapping |
| Key study | Rognmo 2012 among refs 53-57 | Rognmo 2012 (ref 267) | yes — shared |
| Data scale | «currently lacking» | «scarce» | agree: both thin |
| Reading | HIIT not appreciably riskier than MICT | absolute risk low, relative rate 5.6x; more data needed before wide use in CAD | — |
Poon cites the study but reports neither of its rates; both sources call the safety data thin, and they differ in how they weigh the one shared study, so this is a distinction, not a tension (not-joined check i). Franklin’s reading is a type-F refinement of Poon’s paragraph: its safety claim for high-risk groups cites a study whose relative rate was 5.6x, and its next sentence («do not appear to be significantly greater») carries that forward, and the decision depends on whether one weighs the absolute or the relative figure. Small RCTs cannot estimate rare-event rates either way. Practical reading: HIIT stays a reasonable option for the general and overweight adult once active. For known or suspected coronary disease, the statement holds that more safety data are needed before HIIT is widely adopted, especially unsupervised; its default for cardiac patients is regular activity «ideal- ly via traditional medically supervised center- or home- based cardiac rehabilitation programs», and chest pain, lightheadedness or palpitations mean stop and get evaluated -> Exercise-Related Acute Cardiac Events. (Franklin et al., 2020)
For the same head-to-head inside the type-2-diabetes stratum, see HIIT vs Continuous Training for Type 2 Diabetes: a gold MA (Liu 2019, 13 RCTs) finds HIIT beats MICT on VO2peak at moderate certainty (+3.37 ml/kg/min) — concurring with Poon’s direction, though on likely- overlapping trials (not independent type-E) — while its HbA1c/weight edge stays low-certainty.
Brief vigorous exercise snacks may raise CRF (very low certainty). A gold SR+MA (Wan 2025, 14 trials, 483 adults) of short bouts spread across the day (eight trials used bouts of no more than 2 min, six used longer bouts) found snacks significantly improved VO2max (SMD +1.43, 0.61 to 2.25, but VERY LOW certainty and significant only after excluding one high-RoB study; the primary analysis was null) and peak power output (SMD +0.68, 0.00 to 1.36). The PPO gain was confined to physically inactive adults and to bouts longer than 2 min; for VO2max the larger effect in the inactive showed no significant between-group difference. Body composition did not respond, and there are no hard-outcome data — full appraisal and the sufficiency-vs-floor reading: Exercise Snacks and Cardiometabolic Health. (Wan et al., 2025) Reading this as a type-F push of Ross’s «Increases in CRF appear more responsive to increases in intensity than … duration or frequency» principle is the wiki’s link — Wan does not test intensity against duration, and the PPO >2-min condition cuts against ultra-short bouts move CRF. (corrected 2026-10-06: headline distributed 1-2 min snacks still raise CRF / largest in the inactive -> very-low-certainty, PPO-only subgroup limits; type-F link retagged INFERRED; self-critique)
In the obesity stratum, the modality gains don’t separate — and a measurement trap
The dose numbers above come from mostly-fit or mixed populations. Inside the obesity stratum specifically, a gold NMA (O’Donoghue 2021, 45 RCTs, 3566 adults with BMI >=30) ranked all six modalities for CRF and found a sobering result: no modality reached statistical significance. «Although all interventions resulted in an increase in absolute VO2max, there were no statistically significant improvements in fitness found; in accordance with the P score, COM-HI was the intervention … most likely to increase VO2max.» (O’Donoghue et al., 2020) Absolute VO2max did rise 7-15% (COM-HI 15% > AE-V 12.9% > AE-M 9.2% > R-HI 7.4% > R-LM 7.2%), but the between-modality P-score ranking (COM-HI .73 on top) orders non-significant differences — so it establishes no fitness-modality order. This tempers the intensity moves CRF most rule for this stratum: over mostly-short interventions (~20 weeks, 3x45 min/week) in adults with obesity, the CRF gains were real in direction but did not clear significance, and the modalities did not cleanly separate. The full body-composition + CRF ranking lives on Exercise Modality for Body Composition in Obesity.
The measurement trap O’Donoghue flags: report CRF in absolute (L/min), not relative (ml/kg/min), terms in people losing weight. «Reporting a change in relative … as opposed to absolute VO2max … can result in overestimated level of fitness as any loss in body mass automatically increases fitness expressed in relative terms, often without the desired underlying metabolic changes.» (O’Donoghue et al., 2020) A relative-VO2max improvement in a weight-loss study can be arithmetic (smaller denominator), not a true cardiorespiratory gain — a specify-the-measurement-method point for any CRF target in this stratum.
In post-menopausal women, every mode raises CRF — and the modality choice is low-stakes
A second large gold MA supplies the parallel modality question for a different, prevalent stratum: Khalafi 2023 (129 RCTs / 7,141 post-menopausal women overall, age ~53-90, BMI 22-35; the CRF analysis pools 35 arms from 25 studies). Exercise raised CRF with a large overall effect — «Based on the results from 35 intervention arms from 25 studies, exercise training increased CRF (SMD: 1.15; 95% CI: 0.87, 1.42; p = 0.001)» (corrected 2026-10-06: CRF evidence base 129 -> 25 studies; self-critique) — and every mode worked: aerobic SMD 1.21, resistance 1.26, combined 1.47, water-based 0.83 (by-type p-values all significant). (Khalafi et al., 2023) The author framing is explicitly against the ACSM specificity default: «the ACSM recommended combination of aerobic and resistance exercise programs … Nevertheless, our findings revealed that any mode of exercise, including aerobic, resistance, or combined training is effective in improving the CRF in post-menopausal women.» (Khalafi et al., 2023)
Read the by-type ranking as a soft preference, not a result. The subgroup SMDs are reported with p-values but no confidence intervals, so combined’s higher point estimate (1.47) cannot be separated from aerobic (1.21) or resistance (1.26) — all land in the “large” band. The source itself hedges it as «potential greater advantages of combined training.» (Khalafi et al., 2023) And the large overall SMD is likely inflated: I2 = 76.5% (high heterogeneity) and Egger’s test flagged publication bias (p = 0.005, though funnel-plot visual inspection did not). Direction robust; magnitude suspect.
Two measurement limits sit on these numbers. (1) The effects are in SD units (SMD), not ml/kg/min — the authors «were not able to report units of change that were easily interpretable for assessment of clinical meaning» because they pooled across measurement methods. (Khalafi et al., 2023) So the absolute VO2 magnitude is not derivable from this MA (the underivable-optimum discipline — state the interval, not a borrowed absolute). (2) Interval/HIIT training has no CRF estimate here (a prespecified subgroup with too few arms to run) — a gap in the postmenopausal evidence, contrast the HIIT-vs-MICT edge held above for mixed populations.
The cross-stratum synthesis — modality separation for CRF is weak in BOTH strata
Before the claim, the matched-parameter table (NO CELL NO CLAIM):
| Parameter | O’Donoghue 2021 (obesity) | Khalafi 2023 (post-menopausal) | Same quantity? |
|---|---|---|---|
| Population | adults BMI>=30, 76% female | post-menopausal women, BMI 22-35 | NO — different stratum |
| CRF analysis | NMA, P-score rank of absolute VO2max | pairwise SMD vs non-exercise control | NO — rank vs effect size |
| vs-control result | no modality reached significance | all types significant, large SMDs | NO — design differs (CRF study counts similar, 21 vs 25) |
| Top modality for CRF | COM-HI (combined), P .73 | combined, SMD 1.47 (point est.) | YES — same direction |
| Inter-modality separation | none significant | no subgroup CIs -> unconfirmed | YES — not robustly separated |
Two sources with independent authorship and no mutual citation (different populations and designs;
primary-trial overlap unverified — Khalafi’s included-study list is supplementary and not held, and RCTs
in post-menopausal women with obesity could qualify for both) converge on the decision even while
their magnitudes diverge — a convergence on absence of demonstrated separation, which weakly
discriminates modality truly doesn’t matter from both pools were underpowered for between-modality
contrasts. (corrected 2026-10-06: [E-independent] -> overlap unverified; self-critique) combined sits at or near the top for CRF in both, and
in neither do the between-modality CRF differences separate robustly (obesity: nothing significant;
post-menopausal: large effects but no CIs to order them). The magnitude gap — O’Donoghue’s nulls vs
Khalafi’s large SMDs — is explained by stratum, design (the NMA’s indirect comparisons of absolute VO2
vs Khalafi’s direct-vs-control SMDs), and Khalafi’s publication-bias inflation — not by study count,
which is similar for the CRF analyses (Khalafi 25 studies; O’Donoghue’s CRF NMA «Twenty-one studies with
1689 participants» of its 45 RCTs (O’Donoghue et al., 2020);
corrected 2026-10-06: struck power (45 vs 129 RCTs), then matched CRF-analysis to CRF-analysis
counts; self-critique); it is a distinction, not a contradiction
(the “same quantity?” column fails on magnitude, holds on the decision). The beyond-summary payoff:
for CRF, modality choice is low-stakes across both strata — pick for adherence and preference; combined
is a defensible default, and any aerobic-containing mode substantially works. Resistance-only is the
one weak choice, and only for the body-composition outcomes (O’Donoghue), not for CRF.
(Khalafi et al., 2023; inferred from O’Donoghue et al., 2020)
CRF here is the surrogate, and the surrogate->outcome link is cited, not measured. Khalafi motivates the work by «poor CRF is considered an important predictor of all-cause mortality in women» (Khalafi et al., 2023) but measures only exercise->CRF. The CRF->mortality leg is the separately-held cohort/MA evidence (Cardiorespiratory Fitness and Mortality, Kodama) — chaining exercise->CRF->mortality is inference, not this MA’s finding. The strength findings land on Menopause and the Shifting Levers (resistance is the only upper-body lever).
Does raising it help? — the modifiability evidence
This is the partial answer to the nucleus’s predictor-not-lever caveat. Within-person CRF change (a stronger design than cross-sectional comparison) tracks risk:
- Men who went from unfit to fit between two exams had «a reduction in mortality risk of 44%» vs those who stayed unfit (Blair).
- Maintaining or improving CRF gave «27% and 42% reduced risks for CVD and all-cause mortality» (Lee); «Every 1-MET increase in CRF was associated with a 19% lower risk of CVD mortality».
- In «The largest randomized trial of exercise training in HF patients» (HF-ACTION), «every 6% increase in CRF (measured peak V⋅ o2) over 3 months was associated with a 4% lower risk of cardiovascular mortality or cardiovascular hospitalization» «after adjustment for potential confounding variables». Patients were randomized to exercise training, not to a CRF change, so this CRF-change result is an adjusted within-trial association carrying the same confounding risk as the cohorts, in heart-failure patients — not a randomized CRF contrast. (corrected 2026-10-06: the one randomized-trial signal -> adjusted within-trial association; self-critique)
Honest boundary — the upgrade is partial, not complete. The evidence is still overwhelmingly observational; within-person change narrows the reverse-causation worry but does not close it (people whose health is failing get less fit), and no leg is a randomized CRF contrast — HF-ACTION’s CRF-change result is an adjusted within-trial association in heart-failure patients, on a composite that includes hospitalization. The statement’s own framing is calibrated: CRF «is a variable that is responsive to therapy», and improvement «should be communicated to patients» — a modifiable target, not a proven cause of longer life. The proven lever underneath is physical activity (Physical Activity Dose and Mortality); CRF is best read as the trackable, measurable outcome of adherence to that lever.
CRF sharpens a risk estimate — but only if it reclassifies
Adding CRF to a traditional risk model (age, BMI, SBP, diabetes, cholesterol, smoking) improves reclassification, not just correlation: net reclassification improvement (NRI) for CVD mortality of «12.1%» at 10 years in men (Gupta), «27.2%» / «21.0%» in men/women (Stamatakis), «42.8%» in a clinically-referred cohort (Myers). Low CRF flags higher long-term risk within a risk stratum even when short-term risk looks equal — e.g. stage-II hypertension with low vs high CRF carried «18.4% versus 10.1%» 30-year CVD-death risk despite a near-identical «2.3% versus 1.2 %» at 10 years.
The discipline this page must not drop (it is the Risk Modifiers - When Extra Information Changes a Risk Estimate rule): a strong inverse association is not the same as improved prediction. The statement says so — «it does not necessarily mean that CRF directly enhances CVD mortality risk prediction»; «For CRF to truly be a novel risk marker, it must improve risk prediction beyond traditional markers.» The NRI evidence is what earns CRF the reclassification claim; the association alone would not. CRF was nonetheless excluded from the 2013 ACC/AHA risk calculator («it was excluded from the risk calcula- tor») — a gap the statement argues to close. (Ross et al., 2016) Ross’s 2016 statement cannot name the later models, but their own papers show the gap persists. SCORE2: «Original SCORE2 algorithms: Predictors: age, sex, smoking, diabetes, SBP, total and HDL cholesterol» and its diabetes version «Added predictors: age at diabetes diagnosis, HbA1c and eGFR» (SCORE2-Diabetes Working Group and ESC Cardiovascular Risk Collaboration, 2023). PREVENT: «Predictors included traditional risk factors (smoking status, systolic blood pressure, cholesterol, anti-hypertensive or statin use, diabetes) and estimated glomerular filtration rate [eGFR].» (Khan et al., 2024); its optional add-ons are UACR, HbA1c and a social-deprivation index (chunk 01 methods; chunk 02 Table 4), and BMI enters only the heart-failure-specific and death models (chunk 01 methods). Neither includes CRF or any physical-activity measure [searched: fitness/cardiorespiratory/physical activity/exercise across both sources’ chunks, 0 hits; positive control smoking/systolic/eGFR/HbA1c fires in both]. This shows omission, not that adding CRF would improve these models’ discrimination; that needs the NRI-type evidence above, model by model . (corrected 2026-10-06: absent from every current risk model (SCORE2, Framingham, PREVENT) under a Ross tag -> the 2013 calculator exclusion Ross supports; self-critique; SCORE2/PREVENT omission now verified against their own sources)
The guidance move, and how to read it
The statement’s thesis is that CRF should be «an accepted “vital sign”» — the “only major risk factor not routinely assessed in clinical practice”. This is a guidance-layer position from a body (AHA) advocating within its own domain: the 2013 ACC/AHA risk calculator had excluded CRF because the reclassification evidence was then judged inconclusive, and this statement marshals the NRI evidence to argue the exclusion should be revisited. Read it as a well-supported argument, not a neutral guideline — symmetric standards apply to a body making the case for its own risk factor.
Decision relevance
- You can know your CRF for free. An eCRF equation from routine clinical numbers gives a first estimate good enough to identify low fitness; a CPX is only needed for a precise or clinical number. Put it on the FRIEND percentiles to see where you sit.
- From a bike: a test to exhaustion, read through the FRIEND equation. Peak watts from a graded maximal test, converted with the FRIEND equation, avoid the ACSM equation’s average +15% over-read (a cohort mean in the FRIEND registry; the single equation still reads about +3% in women); the per-person error cannot be read from the paper, so treat the number as approximate. For tracking change, repeating the same bike and protocol is likely more informative than the absolute value (constant errors cancel within a person; no held source tests this). (inferred from Kokkinos et al., 2018)
- Without a bike or lab: a timed 2-km walk. For a healthy adult aged 20-65, a fast 2-km walk with a heart-rate strap, read through the sex-specific UKK equation, gives an estimate whose in-sample SEE (9% women, 12-14% men) is about the ±10-15% Ross gives for submaximal tests. Out of sample, in 410 adults aged 18-64, 95% of estimates fell within about ±12.6 ml·kg-1·min-1 (±35%, about ±3.6 METs) of measured VO2max, with agreement worse at higher VO2max; so read it as a band, not a number. Its group mean ran low by 3-12% in Oja’s five groups but not in that sample. Do a practice walk first (Oja) and pace evenly (the authors suggest uneven pacing may explain the walk’s weaker reliability); Oja used telemetry heart rate, and whether a wrist band adds noise is untested. For tracking change, two walk estimates a week apart differed by up to about ±8 ml·kg-1·min-1 in 95% of people, more than the ~10% gain a previously sedentary adult can expect, so one before-and-after pair cannot confirm it; walk time itself repeated within about ±20 s. If you can run, the 20-m shuttle run read tighter on both counts. (Marín-Jiménez et al., 2023; inferred from Oja et al., 1991)
- To raise it: intensity moves it more than duration, and the low-fit gain most from modest activity (the biggest-bang-at-the-low-end rule, consistent across the fitness sources). Expect ~10% from meeting activity guidelines, more from adding intensity or intervals.
- Track CRF as the measurable proxy for the physical-activity lever, not as a separate intervention — the activity is what has the causal warrant; CRF is how you measure whether it is working.
- A low CRF legitimately up-classifies risk within a stratum (a route-(a)/modifier use), and does so on evidence that meets the reclassification bar — but it was excluded from the 2013 ACC/AHA calculator and SCORE2 / SCORE2-Diabetes and PREVENT also omit it (verified against their own papers above), so it informs judgement rather than a computed score. (corrected 2026-10-22: was whether later models include it is unverified.)
- HIIT vs walking for the drifting-median adult — the VO2max edge is real but small at the outcome level (Challenge #11). Intervals raise VO2max more than continuous work (20.6% vs 9.4% above), so HIIT wins the surrogate. But (a) CRF is a predictor, and the mortality dose-response front-loads and flattens (most benefit by ~24 min/day MVPA, Physical Activity Dose and Mortality), so the extra VO2max buys little extra outcome for an under-active person; (b) the advantage is outcome-specific — for the MASLD limb, Fatty Liver MASLD and Weight Loss holds HIIT and moderate-intensity equally effective; and (c) adherence is part of the effect, so a sustained walking habit can beat an abandoned HIIT plan. Poon’s gold umbrella now bounds the surrogate edge itself: HIIT-over-MICT is SMD 0.04–0.64 in healthy adults and SIT-vs-MICT is a wash (0.04–0.18), so even at the surrogate level the interval advantage is modest for an already-normal-fitness adult — the case for HIIT is time-efficiency, not a large fitness win. On the claimed cons: compensation is now held and cuts against the intensity-specific version — Exercise Energy Compensation (Riou 2015) finds compensation real (~18%, up to ~84% long-term) but intensity is not a significant predictor, so HIIT does not compensate more than moderate work; the NEAT-downregulation worry is not HIIT-specific. The affective substrate for the adherence worry is now partly held — Affective Response to High-Intensity Interval Exercise (Niven 2020, gold MA) finds higher-intensity exercise is felt as less pleasant in-task than moderate continuous work (Feeling-Scale MD ~-1.1 during and post), though affect at the end of exercise did not differ (MD -0.72, 95% CI -1.64 to 0.20) and post-exercise enjoyment favoured HIIE — «compared to MICE, HIIE is experienced less positively but post- exercise is reported to be more enjoyable», an inconsistency Niven flags as unresolved for behaviour (corrected 2026-10-06: no post-exercise rebound to rescue it -> enjoyment favoured HIIE; self-critique); and the aversion tracks intensity (HIIE ≈ vigorous continuous), not the interval format, and the step from worse feeling to worse behaviour is explicitly unestablished for HIIE, so worse-HIIT-adherence-than- walking remains unheld as an endpoint (Niven et al., 2020). Net: the case against HIIT for this stratum rests on the flattening mortality curve + adherence, not a compensation penalty. Net for this stratum: doing regular activity at all is the lever; HIIT-vs-walking is a second-order, adherence-bound refinement.
(inferred from Ross et al., 2016)
Limits
- Expert-consensus scientific statement, not a systematic review or GRADE appraisal — «not intended to be a comprehensive review». Its evidence base is overwhelmingly observational.
- Causality on hard outcomes is inferred, not established (HF-ACTION’s CRF-change result is an adjusted within-trial association, not a randomized contrast; corrected 2026-10-06: was HF-ACTION aside; self-critique) — the modifiability finding upgrades but does not resolve the nucleus’s predictor caveat.
- ~50% of CRF is heritable — the trainable fraction is real but bounded.
- One body (AHA), 2016; whether other bodies endorse the vital-sign framing is unprobed.
(inferred from Ross et al., 2016)