Comparison of objectively measured and estimated cardiorespiratory fitness to predict all-cause and cardiovascular disease mortality in adults: A systematic review and meta-analysis of 42 studies representing 35 cohorts and 3.8 million observations

Highlights

• •We meta-analyzed 42 studies representing 35 cohorts and 3.8 million adults to compare the associations of objectively measured, exercise-estimated, and non-exercise-estimated cardiorespiratory fitness (CRF) with all-cause and cardiovascular disease (CVD) mortality in adults.
• •We found 14% and 16% reductions in all-cause and CVD mortality risk per higher metabolic equivalent of task ((MET) i.e., 3.5 mL/kg/min), respectively, with no differences in risk reduction between objectively measured, exercise-estimated, and non-exercise-estimated CRF.
• •Exercise and non-exercise estimated CRF provide practical and robust alternatives to the more costly and time-consuming objectively measured CRF to enhance patient risk stratification in clinical settings.

Abstract

Background

Cardiorespiratory fitness (CRF) is a powerful health marker recommended by the American Heart Association as a clinical vital sign. Comparing the predictive validity of objectively measured CRF (the “gold standard”) and estimated CRF is clinically relevant because estimated CRF is more feasible. Our objective was to meta-analyze cohort studies to compare the associations of objectively measured, exercise-estimated, and non-exercise-estimated CRF with all-cause and cardiovascular disease (CVD) mortality in adults.

Methods

Systematic searches were conducted in 9 databases (MEDLINE, SPORTDiscus, Embase, Scopus, PsycINFO, Web of Science, PubMed, CINAHL, and the Cochrane Library) up to April 11, 2024. We included full-text refereed cohort studies published in English that quantified the association (using risk estimates with 95% confidence intervals (95%CIs)) of objectively measured, exercise-estimated, and non-exercise-estimated CRF with all-cause and CVD mortality in adults. CRF was expressed as metabolic equivalents (METs) of task. Pooled relative risks (RR) for all-cause and CVD mortality per 1-MET (3.5 mL/kg/min) higher level of CRF were quantified using random-effects models.

Results

Forty-two studies representing 35 cohorts and 3,813,484 observations (81% male) (362,771 all-cause and 56,471 CVD deaths) were included. The pooled RRs for all-cause and CVD mortality per higher MET were 0.86 (95%CI: 0.83–0.88) and 0.84 (95%CI: 0.80–0.87), respectively. For both all-cause and CVD mortality, there were no statistically significant differences in RR per higher MET between objectively measured (RR range: 0.86–0.90) and maximal exercise-estimated (RR range: 0.85–0.86), submaximal exercise-estimated (RR range: 0.91–0.94), and non-exercise-estimated CRF (RR range: 0.81–0.85).

Conclusion

Objectively measured and estimated CRF showed similar dose–response associations for all-cause and CVD mortality in adults. Estimated CRF could provide a practical and robust alternative to objectively measured CRF for assessing mortality risk across diverse populations. Our findings underscore the health-related benefits of higher CRF and advocate for its integration into clinical practice to enhance risk stratification.

Graphical abstract

1. Introduction

Cardiorespiratory fitness (CRF) reflects the capacity of the physiological systems to perform whole-body, dynamic, sustained physical activity.¹ Maximal oxygen uptake (VO_2max), the highest rate of oxygen consumption during maximal aerobic exercise, is considered the best measure of CRF.² Expressed per kilogram of body mass (mL/kg/min) or as metabolic equivalents (METs) of task, VO_2max can be objectively measured in laboratory settings using indirect calorimetry (also called cardiopulmonary exercise testing (CPET)). When this “gold standard” measurement is not possible, VO_2max can be estimated using validated exercise tests conducted in laboratory or field settings without CPET.³ Different exercise types (e.g., treadmill running/walking, cycle ergometry) can be utilized. A growing number of non-exercise algorithms based on health indicators such as age, body mass index (BMI), and physical activity level are also used to estimate VO_2max.^2,4

CRF is recommended by the American Heart Association as a clinical vital sign.² Meta-analyses of cohort studies indicate that adult CRF is inversely associated with all-cause mortality,5, 6, 7, 8 cardiovascular disease (CVD) mortality,^7,8 cancer mortality,⁷ incident heart failure,⁹ stroke,¹⁰ some cancers,¹¹ type 2 diabetes,¹² hypertension,¹³ dementia,¹⁴ and depression.¹⁵ When added to traditional risk factors, CRF significantly improves the reclassification of risk for adverse outcomes and risk stratification.² Compared to other modifiable risk factors such as smoking and obesity, CRF is the strongest independent predictor of health and longevity.^2,16,17

The clinical utility of CRF has mostly been evaluated through meta-analyses of its predictive validity.5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 Kodama et al.⁵ initially reported that a 1-MET (3.5 mL/kg/min) higher level of CRF corresponded to 13% and 15% reductions in all-cause mortality and CVD events, respectively. This reduction in mortality risk was progressively lower with increased CRF.^5.By consolidating findings for objectively measured and exercise-estimated CRF, Laukkanen et al.⁶ and Han et al.⁷ found risk reductions of 11%–12% per higher MET for all-cause mortality. Han et al.⁷ also found a 13% risk reduction in CVD mortality per higher MET. Furthermore, Qiu et al.⁸ found a 17% risk reduction per higher MET of non-exercise-estimated CRF for all-cause and CVD mortality.

However, a critical gap persists: no meta-analysis has methodically compared the association of objectively measured CRF (the “gold standard”) against estimated variants with all-cause and CVD mortality in adults. To update and extend the meta-analyses of Laukkanen et al.,⁶ Han et al.,⁷ and Qiu et al.,⁸ we meta-analyzed cohort studies to compare the association of objectively measured CRF to that of exercise-estimated (maximal/submaximal) and non-exercise-estimated CRF with all-cause and CVD mortality in adults. In doing so, we aim to provide valuable insights that will shape future research, help guide policies, and potentially improve clinical practice in primary and secondary prevention of disease.

2. Methods

2.1. Registration and protocol

This systematic review and meta-analysis protocol was prospectively registered with PROSPERO on the March 31, 2022 (Trial registration ID: CRD42022306213; accessed: https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=306213). We followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 statement (Supplementary Table 1).¹⁸

2.2. Eligibility criteria

• (1)Population: General adult population aged ≥ 18 years at baseline. Studies on participants with prevalent chronic conditions at baseline (e.g., coronary heart disease, diabetes, hypertension) were excluded.
• (2)Exposure: Objectively measured, exercise-estimated, or non-exercise-estimated CRF at baseline expressed as VO_2max in mL/kg/min or METs. Objectively measured CRF means measured VO_2max from maximal aerobic exercise testing using CPET; exercise-estimated CRF studies means estimated VO_2max from a maximal/submaximal exercise work/heart rate; and non-exercise-estimated CRF means estimated VO_2max from baseline non-exercise variables (e.g., age, sex, BMI, waist circumference, physical activity level, resting heart rate, smoking status) using a prediction equation. We classified exercise-estimated CRF by exercise intensity as maximal (symptom-limited exercise stress test to volitional fatigue or to an intensity ≥ 85% age-predicted maximal heart rate) or submaximal (intensity < 85% age-predicted maximal heart rate). Studies were included if they reported the association of CRF with all-cause or CVD mortality. Risk estimates such as relative risks (RRs) or hazard ratios (HRs) with 95% confidence intervals (95%CIs) must have been reported for CRF per incremental increase or across at least 3 levels.
• (3)Outcome: All-cause or CVD mortality.
• (4)Study design: Cohort (prospective and retrospective). Studies must have reported on unique cohorts. When multiple studies were available for the same cohort investigating the same outcomes, we included the most recently published study with the most complete coverage (baseline sample size and span of measurement years) and longest follow-up. Other study designs (e.g., randomized controlled trial (RCT), experimental, case-control, cross-sectional, review, qualitative, non-empirical) were excluded.
• (5)Publication status: Full-text refereed journal articles published in English. Conference abstracts/papers, commentaries, editorials, dissertations, and grey literature were excluded.

2.3. Information sources and search strategy

We updated the Laukkanen et al.,⁶ Han et al.,⁷ and Qiu et al.⁸ meta-analyses by searching 9 databases (MEDLINE (via Ovid), SPORTDiscus (via EBSCOhost), Embase (via Ovid), Scopus, PsycINFO, Web of Science, PubMed (via Ovid MEDLINE), CINAHL (via EBSCOhost), and the Cochrane Library) for cohort studies from December 26, 2019 (the earliest search date of the 3 meta-analyses) to January 25, 2023. The search was later updated from January 1, 2023 to April 11, 2024. The search strategy for databases involved terms related to the following keywords: cardiorespiratory fitness, mortality, cardiovascular disease, and cohort studies. Search strategies for databases are shown in Supplementary Table 2. We included unique cohort studies that were included in the Laukkanen et al.,⁶ Han et al.,⁷ and Qiu et al.⁸ meta-analyses, as well as those published thereafter. We also searched the reference lists of included studies and topical reviews.

2.4. Selection process

Records were imported into EndNote (V7.7.1; Clarivate Analytics, Philadelphia, PA, USA) and de-duplicated, and then into Covidence (Veritas Health Innovation, Melbourne, VIC, Australia) for further de-duplication and record screening. Titles and abstracts were independently screened against inclusion criteria by two of the following authors with vast experience conducting and publishing systematic reviews (BS, CCS, BGGC, JPC, JJL, BG, and GRT). Full-text studies were then independently screened against inclusion criteria by the same authors. A third author resolved conflicts.

2.5. Data collection process and data items

Data were independently extracted using a pre-designed Excel spreadsheet (Microsoft, Redmond, WA, USA) by two of the following authors (BS, CCS, and BGGC). A third author resolved conflicts. The following data were extracted: first author's surname, publication year, country of origin, sample size, number of events, follow-up duration, gender distribution, mean/median age of participants at baseline, method of CRF assessment, CRF levels, number of events, person-years, and the number of participants per CRF level. Additionally, the fully adjusted RRs or HRs with corresponding 95%CIs, plus the list of covariates used for statistical adjustment, were extracted. If required, we requested additional information from corresponding authors by email to clarify study details and to avoid including data from two studies using the same cohort.

2.6. Quality assessment

Study quality was assessed by the Newcastle-Ottawa Scale (NOS) for cohort studies.¹⁹ Briefly, the NOS is an 8-item scale that assesses selection, comparability, and outcome. This scale assigns scores ranging from 0 to 9, with higher scores indicating better study quality. Studies were categorized as poor (0–3 points), fair (4–6 points), or good (7–9 points) based on their overall scores. Study quality was independently assessed by two of the following authors (BS, CCS, BGGC, BG, and GRT), with conflicts resolved by a third author.

2.7. Data synthesis and analysis

A random-effects model was used to pool RRs and 95%CIs for dose–response and categorical meta-analyses. RRs (95%CIs) were used as the effect statistic, as we assumed that HRs reported in the primary studies approximated the RRs. We visualized the study-specific and pooled RRs (95%CIs) using forest plots. A dose–response meta-analysis was performed to quantify risk reduction in all-cause and CVD mortality per higher MET of CRF. Pairwise comparisons (Q_b) were performed to assess differences in associations between objectively measured CRF and (1) maximal exercise-estimated CRF; (2) submaximal exercise-estimated CRF; and (3) non-exercise-estimated CRF. If missing, we used Greenland and Longnecker's²⁰ method to estimate study-specific RRs per higher MET of CRF. Categorical meta-analyses were performed by assigning study-specific RRs to either the lowest vs. highest CRF or the lowest vs. intermediate CRF. When analysing studies involving more than 3 CRF groups, the intermediate category was determined by aggregating all medium groups. RRs for the intermediate category were calculated using either a random-effects or fixed-effects model. We standardized all reported categorical RRs as the risk of the lowest CRF level relative to the higher CRF level. METs were used as the common CRF metric for all analyses. We converted all CRF values into METs using an approach described elsewhere.⁵ To avoid overlap across the 3 CRF levels, we defined the CRF levels as lowest (<7.9 METs), intermediate (7.9–10.8 METs), and highest (>10.8 METs), as per Kodama et al.⁵ The findings from our dose–response meta-analyses were primary; findings from the categorical meta-analyses were supplementary and were not discussed.

Statistical heterogeneity was assessed using Q and I² statistics, with I² values interpreted as negligible (I²: 0%–40%), moderate (I²: 30%–60%), substantial (I²: 50%–90%), or considerable (I²: 75%–100%).²¹ We examined potential sources of heterogeneity by performing the following subgroup analyses using the Q test based on analysis of variance: publication year (>2015 and ≤2015), geographical region (Asia, Europe, North America, the Caribbean, and the Middle East), gender (men, women, and mixed), age (>50 years, ≤50 years, and not reported), follow-up duration (>10 years and ≤10 years), number of all-cause/CVD mortality events (>600 and ≤600), and exercise type (cycle ergometry, treadmill (running/walking), bench stepping, and non-exercise algorithms). Publication bias was assessed visually by funnel plots. All statistical analyses were conducted using Stata/MP (V16.0; Stata Corp., College Station, TX, USA). We used an α level of 0.05.

2.8. Certainty assessment

The levels of evidence and grades for recommendations from the Oxford Centre for Evidence-Based Medicine²² were used to categorize the overall level of evidence as: Grade A: consistent Level 1 studies (i.e., systematic reviews of RCTs or individual RCTs); Grade B: consistent Level 2 studies (i.e., systematic reviews of cohort studies or individual cohort studies), Level 3 studies (i.e., systematic reviews of case-control studies or individual case-control studies), or extrapolations from Level 1 studies; Grade C: Level 4 studies (i.e., case series or extrapolations from Levels 2 or 3 studies); or Grade D: Level 5 evidence (i.e., expert opinion without explicit critical appraisal) or studies that are troublingly inconsistent or inconclusive at any level.

2.9. Deviation from registered protocol

We planned a subgroup analysis based on CVD risk status; however, it was impossible due to the way in which data were presented across the primary studies. We added subgroup analyses based on the year of publication and number of mortality events since they provided additional insights into the CRF-mortality associations.

3. Results

3.1. Study selection

Fig. 1 presents a flow diagram of the literature search and screening process. Following de-duplication, 3929 records were identified from database searching, with 69 additional records identified from other sources. Forty-two studies23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64 representing 35 unique cohorts were included.

Fig. 1.

Flow diagram showing the flow of studies through different phases of the systematic review.

3.2. Study characteristics and quality assessment

Study characteristics are summarized in Table 1. Cohorts were from North America (n = 15),^23,25, 26, 27, 28^,35, 36, 37, 38^,42,54,58,61, 62, 63 Europe (n = 14),29, 30, 31^,39, 40, 41^{,46,47,49,52,55,57,59,60} Asia (n = 4),^{45, 53,56,64} the Caribbean (n = 1),⁵⁰ and the Middle-East (n = 1).²⁴ There were 3,813,484 observations (81% male) across all 42 studies (n = 362,771 all-cause and 56,471 CVD mortality events) and a median sample size of 13,887 (interquartile range (IQR): 4137–38,480). Median participant age was 52 years (IQR: 46–56 years) and the median follow-up duration was 9 years (IQR: 7–18 years). Of the 31 studies that objectively measured (n = 5 (16%)^{38,39,43,46,50}) or estimated (n = 26 (84%)23, 24, 25^,27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37^,40, 41, 42^,47,48,53, 54, 55^,58, 59, 60^,62) CRF using exercise, most used maximal exercise (n = 26 (84%)23, 24, 25^,27, 28, 29^,31, 32, 33, 34, 35, 36, 37, 38, 39^,42,43,46, 47, 48^,50,53, 54, 55^,58,59) and treadmill running/walking (n = 18 (58%)23, 24, 25^,27,28,32, 33, 34, 35, 36, 37, 38^{,42,46,48,54,58,59}). The outcome measures were all-cause mortality (n = 38 (90%)23, 24, 25, 26^,29, 30, 31^,33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51^,53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64) and CVD mortality (n = 27 (64%)26, 27, 28, 29, 30, 31, 32, 33^{,36,38,39,41,43,}45, 46, 47^,50, 51, 52, 53^,55,57, 58, 59^,61,63,64). Study quality was graded as good (median = 9; IQR: 8–9; Supplementary Table 3).

Table 1.

Summary of the 42 included studies.

Study	Country	Sample size (%male)	Mean/median age (SD/range) (year)	Follow-up (range) (year)	Cardiorespiratory fitness assessment	Exercise intensity	Specific outcomes: number of deaths
Aijaz et al. (2008)23	USA	8620 (73%)	52.0	16.0	Treadmill (Bruce)	Maximal	All-cause mortality: 535
Aker et al. (2023)24	Israel	6836 (56%)	51.9 (5.9)	7.1	Treadmill (Bruce)	Maximal	All-cause mortality: 118
Al-Mallah et al. (2016)25	USA	57,284 (52%)	53.4	10.0	Treadmill (Bruce)	Maximal	All-cause mortality: 6402
Artero et al. (2014)26a	USA	43,356 (79%)	44.5	14.5	Algorithmg	NA	All-cause mortality: 1933 CVD mortality: 627
Balady et al. (2004)27b	USA	3043 (47%)	45.0	18.2	Treadmill (Bruce)	Maximal	CVD mortality: 305
Bruce et al. (1980)28	USA	2365 (100%)	45.0	5.6	Treadmill (Bruce)	Maximal	CVD mortality: 47
Crump et al. (2017)29	Sweden	1,547,478 (100%)	18.0	28.2	Cycle ergometer (Ramp)	Maximal	All-cause mortality: 64,343 CVD mortality: 10,381
Ekblom-Bak et al. (2019)30	Sweden	266,109 (53%)	(18.0–74.0)	7.6	Cycle ergometer (Åstrand)	Submaximal	All-cause mortality: 2750 CVD mortality: 455
Engeseth et al. (2018)31	Norway	2014 (100%)	50.0 (40.0–59.0)	35.0	Cycle ergometer (Ramp)	Maximal	All-cause mortality: 1178 CVD mortality: 528
Farrell et al. (2020)32a	USA	19,838 (0%)	44.9 (10.5)	19.2	Treadmill (Modified Balke)	Maximal	CVD mortality: 391
Farrell et al. (2020)33a	USA	Cohort 1: 24,475 (100%) Cohort 2: 23,387 (100%)	Cohort 1: 42.0 Cohort 2: 45.4	Cohort 1: 10.5 Cohort 2: 12.1	Treadmill (Modified Balke)	Maximal	All-cause mortality: Cohort 1: 646 Cohort 2: 526 CVD mortality: Cohort 1: 190 Cohort 2: 109
Farrell et al. (2022)34a	USA	17,901 (0%)	45.9	17.9	Treadmill (Modified Balke)	Maximal	All-cause mortality: 1198
Goraya et al. (2000)35	USA	3107 (60%)	48.6	6.3	Treadmill (Bruce, modified Bruce, or Naughton)	Maximal	All-cause mortality: 224
Gulati et al. (2005)36	USA	5636 (0%)	52.0	9.0	Treadmill (Bruce)	Maximal	All-cause mortality: 171 CVD mortality: 52
Harb et al. (2021)37c	USA	126,356 (59%)	53.5 (12.6)	8.7	Treadmill (Bruce, modified Bruce, Cornell (0, 5, 10), Naughton, modified Naughton)	Maximal	All-cause mortality: 9929
Imboden et al. (2018)38	USA	4137 (56%)	42.8	24.2	Treadmill (CPET, gas: Bruce, Ball State University Bruce Ramp, modified Balke-Ware)	Maximal	All-cause mortality: 727 CVD mortality: 212
Jae et al. (2021)39	Finland	2368 (100%)	52.9	25.0 (18.0–27.0)	Cycle ergometer (CPET, gas, Ramp)	Maximal	All-cause mortality: 1116 CVD mortality: 512
Jensen et al. (2017)40	Denmark	5131 (100%)	48.8 (5.4)	28.3	Cycle ergometer (Åstrand)	Submaximal	All-cause mortality: 4486
Kim et al. (2018)41	UK	70,913 (47%)	57.2 (8.2)	5.7	Cycle ergometer (Ramp)	Submaximal	All-cause mortality: 832 CVD mortality: 117
Kokkinos et al. (2022)42	USA	750,302 (94%)	61.9	10.2	Treadmill (Bruce)	Maximal	All-cause mortality: 174,807
Laukkanen et al. (2021)43	Finland	2205 (100%)	52.8	28.5	Cycle ergometer (CPET, gas, Ramp)	Maximal	All-cause mortality: 1348 CVD mortality: 607
Lee et al. (2021)44b	USA	2962 (100%)	60.6	15.0	Algorithm	NA	All-cause mortality: 770
Lee et al. (2022)45	The Republic of Korea	38,350 (43%)	53.0	7.3	Algorithm	NA	All-cause mortality: 1474 CVD mortality: 325
Letnes et al. (2019)46d	Norway	4527 (49%)	48.2	8.8	Treadmill (CPET, gas)	Maximal	All-cause mortality: 91 CVD mortality: 18
Lindow et al. (2020)47	Sweden	13,887 (54%)	59.7	7.7 (5.6–11.4)	Cycle ergometer (Ramp)	Maximal	All-cause mortality: 1809 CVD mortality: 546
Mandsager et al. (2018)48c	USA	122,007 (59%)	53.4 (12.6)	8.4	Treadmill (Bruce, modified Bruce, Cornell (0, 5, 10), Naughton, modified Naughton)	Maximal	All-cause mortality: 13,637
Martinez-Gomez et al. (2015)49	Spain	2930 (50%)	Men: 70.2 Women: 68.6	9.4	Algorithm	NA	All-cause mortality: 865
Miller et al. (2005)50	Trinidad and Tobago	578 (100%)	52.0	7.3	Cycle ergometer (CPET, gas, Ramp)	Maximal	All-cause mortality: 68 CVD mortality: 62
Nauman et al. (2017)51d	Norway	38,480 (49%)	48.3	16.3	Algorithm	NA	All-cause mortality: 3863 CVD mortality: 1133
Nes et al. (2014)52d	Norway	37,112 (49%)	Men: 43.5 Women: 44.1	24.0	Algorithm	NA	CVD mortality: 13,453
Park et al. (2009)53	The Republic of Korea	18,775 (100%)	55.7 (11.1)	6.4	Cycle ergometer (Ramp)	Maximal	All-cause mortality: 547 CVD mortality: 101
Phan et al. (2022)54	USA	40,520 (49%)	68.8	3.5	Treadmill	Maximal	All-cause mortality: 961
Salokari et al. (2019)55e	Finland	3758 (60%)	55.8	6.3	Cycle ergometer (Ramp)	Maximal	All-cause mortality: 404 CVD mortality: 3758
Song et al. (2019)56	The Republic of Korea	14,122 (43%)	Men: 69.1 Women: 70.4	3.0	Algorithm	NA	All-cause mortality: 701
Stamatakis et al. (2013)57	UK	32,319 (45%)	Men: 50.4 Women: 50.7	9.0	Algorithm	NA	All-cause mortality: 2165 CVD mortality: 460
Stevens et al. (2002)58	USA	5366 (53%)	46.0	NR	Treadmill (Bruce)	Maximal	All-cause mortality: 1166 CVD mortality: 449
Stevens et al. (2004)59f	Russia	1359 (100%)	49.0	17.6	Treadmill (Bruce)	Maximal	All-cause mortality: 211 CVD mortality: 98
Tarp et al. (2021)60	UK	77,169 (48%)	57.8	7.4	Cycle ergometer (Ramp)	Submaximal	All-cause mortality: 1731
Vainshelboim et al. (2022)61	USA	330,769 (56%)	61.1	14.9	Algorithm	NA	All-cause mortality: 54,612 CVD mortality: 20,177
Villeneuve et al. (1998)62	Canada	7561 (48%)	45.0	7.0	Bench stepping (Canadian Aerobic Fitness Test)	Submaximal	All-cause mortality: 129
Zhang et al. (2017)63	USA	12,506 (49%)	Men: 43.4 Women: 44.0	19.2	Algorithm	NA	All-cause mortality: 3439 CVD mortality: 999
Zhao et al. (2022)64	China	15,566 (39%)	51.5	6.0	Algorithm	NA	All-cause mortality: 859 CVD mortality: 359

^aData were from the Aerobics Center Longitudinal Study (ACLS) or Cooper Center Longitudinal Study (CCLS).

^bData were from the Framingham Heart Study.

^cData were from the Cleveland Clinic.

^dData were from the Nord-Trøndelag Health Study (HUNT).

^eWe only included data on the Finnish Cardiovascular Study (FINCAVAS) not the Kuopio Ischaemic Heart Disease study (KIHD)—data on the KIHD cohort were included in Laukkanen et al. (2021).⁴³

^fWe only included data on Russian men, not U.S. men, from the Lipids Research Clinics First Prevalence Studies—data on U.S. men and women were included in Stevens et al. (2002).⁵⁸

^gRisk estimates were available for both maximal exercise-estimated CRF and non-exercise-estimated CRF, with only non-exercise-estimated CRF data included because maximal exercise-estimated CRF data were included in other studies that used the ACLS or CCLS (Farrell et al. (2020),³² Farrell et al. (2020),³³ and Farrell et al. (2022)³⁴).

Abbreviations: CPET = cardiopulmonary exercise testing; CVD = cardiovascular disease; NA = not applicable; NR = not reported.

3.3. Synthesis of results

3.3.1. CRF and risk of all-cause mortality

Fig. 2 shows the dose–response meta-analysis of all-cause mortality per 1-MET higher level of CRF (n = 1,813,565 participants and 273,645 events for the dose–response analysis, i.e., excluding the categorical data). The pooled RR for all-cause mortality per higher MET was 0.86 (95%CI: 0.83–0.88; I² = 97.66%; Q = 1893.49; p < 0.01). There was considerable between-study heterogeneity, but no evidence of publication bias. There were no statistically significant differences in pooled RRs for all-cause mortality per higher MET between objectively measured CRF (RR = 0.90, 95%CI: 0.83–0.98) and maximal exercise-estimated CRF (RR = 0.85, 95%CI: 0.81–0.88); test of group differences (Q_b) = 1.94; p = 0.16), submaximal exercise-estimated CRF (RR = 0.94, 95%CI: 0.90–0.98); Q_b = 0.80; p = 0.37), and non-exercise-estimated CRF (RR = 0.85, 95%CI: 0.80–0.89); Q_b = 1.69; p = 0.19).

Fig. 2.

Meta-analysis of all-cause mortality per 1-metabolic equivalent higher level of CRF. 95%CI = 95% confidence interval; CRF = cardiorespiratory fitness.

Results of subgroup analyses of all-cause mortality per higher MET are in Fig. 3. Statistically significant risk reduction in all-cause mortality per higher MET was found in all subgroups, except exercise type (bench stepping). Dose–response associations differed by geographical region (Q_b = 18.61; p < 0.01). Supplementary Figs. 1 and 2 show the categorical meta-analyses of all-cause mortality for adults with lowest vs. highest CRF and lowest vs. intermediate CRF, respectively.

Fig. 3.

Subgroup analysis of all-cause mortality per 1-metabolic equivalent higher level of cardiorespiratory fitness. 95%CI = 95%confidence interval.

3.3.2. CRF and risk of CVD mortality

The dose–response meta-analysis of CVD mortality per higher MET is shown in Fig. 4 (n = 870,874 participants and 39,720 events). The pooled RR for CVD mortality per higher MET was 0.84 (95%CI: 0.80–0.87; I² = 92.50%; Q_b = 231.48; p < 0.01). Between-study heterogeneity was considerable and there was no evidence of publication bias. There were no statistically significant differences in pooled RRs for CVD mortality per higher MET between objectively measured CRF (RR = 0.86, 95%CI: 0.78–0.95) and maximal exercise-estimated CRF (RR = 0.86, 95%CI: 0.81–0.91); Q_b = 0.00; p = 0.98, submaximal exercise-estimated CRF (RR = 0.91, 95%CI: 0.90–0.93; Q_b = 1.10; p = 0.29), and non-exercise-estimated CRF (RR = 0.81, 95%CI: 0.76–0.86); Q_b = 0.94; p = 0.33).

Fig. 4.

Meta-analysis of cardiovascular disease mortality per 1-metabolic equivalent higher level of CRF. 95%CI = 95% confidence interval; CRF = cardiorespiratory fitness.

Fig. 5 shows the results of subgroup analyses of CVD mortality per higher MET. Risk reduction in CVD mortality per higher MET was consistently found in all subgroups, with a stronger association (Q_b = 5.58, p = 0.02) for shorter follow-up durations (≤10 years: RR = 0.78, 95%CI: 0.70–0.85) compared to longer follow-up durations (>10 years: RR = 0.87, 95%CI: 0.84–0.89). Supplementary Figs. 3 and 4 show the categorical meta-analyses of CVD mortality for adults with lowest vs. highest CRF and lowest vs. intermediate CRF, respectively.

Fig. 5.

Subgroup analysis of cardiovascular disease mortality per 1-metabolic equivalent higher level of cardiorespiratory fitness. 95%CI = 95% confidence interval.

3.4. Certainty of evidence

The association between CRF and all-cause mortality was supported by a Grade B recommendation, based on consistent findings from Level 2 (cohort) studies. For this reason, the association between CRF and CVD mortality was supported by a Grade B recommendation.

4. Discussion

We compared the associations of objectively measured CRF and both exercise-estimated and non-exercise-estimated CRF with all-cause and CVD mortality in adults. We found a 14% and a 16% reduction in all-cause and CVD mortality risk per higher MET, respectively. Dose–response associations were similar regardless of CRF assessment method. These findings are clinically important as they suggest that the easily assessed exercise and non-exercise estimates of CRF are similarly associated with all-cause and CVD mortality as the more costly and time-consuming objectively measured CRF (the “gold standard”). Estimated CRF could be implemented in clinical settings when objectively measured CRF is not feasible without losing predictive utility for all-cause and CVD mortality.

Published meta-analyses have found notable risk reductions for all-cause and CVD mortality per higher MET.6, 7, 8 Our findings extend those of others6, 7, 8 by including new data from 13 cohort studies; ^24,34,37,39,42, 43, 44, 45^{,47,54,60,61,64} 696,117 observations; 75,213 all-cause deaths; and 26,163 CVD deaths. Compared to others,6, 7, 8 we found similar risk reductions in all-cause and CVD mortality per higher MET of objectively measured, exercise-estimated, and non-exercise-estimated CRF. We performed subgroup analyses of potentially relevant study-level characteristics to examine between-study heterogeneity. The finding of risk reduction in all-cause and CVD mortality per higher MET was consistent across almost all subgroups. The degree of risk reduction differed by geographical region and age for all-cause mortality, and follow-up duration, gender, and age for CVD mortality. Our findings are similar to others^5,6 who found that studies with longer follow-up durations had weaker associations compared to those with shorter follow-up durations. These findings may be due to regression dilution bias. Nevertheless, single CRF assessments remain robust predictors of mortality risk for at least 10 years. This aligns with the commonly used timescale in validated CVD risk calculators.⁶ That is, initial CRF assessment can provide valuable information about mortality risk with a similar timescale as traditional risk factors such as blood pressure and cholesterol.

Compared to objectively measured CRF, exercise-estimated CRF was assessed in 6.5-fold more cohorts representing 234-fold more participants. Notably, both assessment methods were similarly associated with mortality, suggesting that exercise-estimated CRF could provide a practical and feasible alternative to objectively measured CRF. Unlike objectively measured CRF, which requires costly and intricate CPET equipment and trained personnel for administration and interpretation, exercise-estimated CRF methods are more accessible and readily implemented across diverse clinical and research environments, potentially benefiting a wider population. Given similar associations for submaximal exercise-estimated and objectively measured CRF, our findings suggest that submaximal testing offers a viable health-related alternative. This assumption might be clinically important or preferred by clinicians because the risk of adverse events is lower for submaximal exercise compared to maximal exercise.⁶⁵ Lowered risk of adverse events is beneficial, for example, in individuals deemed to be at high risk or with reduced exercise tolerance.⁶⁶ Moreover, submaximal testing is simple, cost-effective, and efficient, making it a practical and appealing option for widespread clinical and research use. Whenever possible, either objectively measured or exercise-estimated CRF are recommended for clinical application; however, clinicians do have the option of using non-exercise-estimated CRF measures to inform their practice. The strength of non-exercise-estimated CRF is clinically important given the call for CRF to be routinely applied as a risk factor² and the fact that exercise testing is not yet regularly administered by primary health professionals or recommended for asymptomatic individuals due to its limited predictive utility.⁶⁷ However, certain factors may limit the clinical utility of non-exercise-estimated CRF, such as the responsiveness of non-exercise CRF to intervention. Understanding the precision and adaptability of non-exercise CRF is vital for its successful integration into clinical settings and for use by clinicians.

Understanding the mechanisms involved in CRF sheds light on its significance in health assessment. Higher CRF is associated with multiple potential protective mechanisms, including improved cardiovascular function, better tissue oxygen delivery, reduced blood pressure, adiposity, and chronic inflammation, and improved glucose metabolism and lipid profiles.68, 69, 70, 71 Collectively, these factors reduce the risk of CVD and positively influence overall health.² Including CRF in routine practice affords clinicians a vitally important opportunity to improve individuals’ health levels by enhancing risk stratification and better informing treatment/management strategies.²

5. Strengths and limitations

Our study has several strengths. First, we believe this is the first study to systematically compare the associations of objectively measured and estimated CRF with all-cause and CVD mortality in adults. Second, we used a rigorous systematic review strategy and conducted high-quality and extensive meta-analyses. Importantly, our meta-analysis encompassed a substantial dataset of 42 studies representing 35 cohorts and 3.8 million adults with 362,771 all-cause and 56,471 CVD deaths, which contributes to the robustness of our conclusions. We also graded the overall certainty of evidence.

There were also study limitations. Including only English language studies may have meant we missed relevant studies published in other languages. Evidence suggests that limiting systematic reviews to English language studies minimally impacts effect estimates and conclusions.^72,73 Moreover, the considerable between-study heterogeneity indicates that our findings should be interpreted with caution. Our meta-analysis was also based on single measures/estimates of CRF, but not change measures/estimates of CRF, which meant we could not examine whether improved CRF reduced all-cause or CVD mortality risk. Although we extracted fully adjusted risk estimates, it is possible that our findings were biased by residual/unmeasured confounding. Because no study measured and estimated CRF within the same cohort, and without individual-level data, neither direct head-to-head comparisons nor indirect comparisons from unobserved head-to-head comparisons using a common comparator could be made.

6. Conclusion

Our meta-analysis underscores the importance of CRF as a key predictor of all-cause and CVD mortality. Estimated CRF was similarly associated with all-cause and CVD mortality as the “gold standard” objectively measured CRF, and therefore our results confirm the predictive utility of non-exercise estimates of CRF when exercise testing is unavailable. Exercise-estimated CRF represents a popular and efficient alternative to objective measurement, broadening its applicability in diverse settings. The consistency of the mortality risk reduction across varied subgroups reinforces the robustness of CRF as an indicator of mortality risk. These insights highlight the inherent health benefits of higher CRF and underscore the need for its integration into clinical practice for refined risk stratification. To confirm our findings and before implementing estimated CRF in clinical setting when objectively measured CRF is not feasible, future research should compare associations of objectively measured CRF and estimated CRF with other clinical outcomes (e.g., comparing their validity for predicting incident morbidity).

Authors’ contributions

BS acquired and interpreted the data, statistically analyzed the data, and drafted the manuscript; CCS acquired and interpreted the data and drafted the manuscript; JCP, JPC, MCG, FBO, JJL and RM conceptualized and designed the study, acquired and interpreted the data, and reviewed and edited the manuscript for important intellectual content; BGGC, CM, NMJ, PMG, JM, and BG acquired and interpreted the data as well as reviewed and edited the manuscript for important intellectual content; GRT conceptualized and designed the study, acquired and interpreted the data, and drafted the manuscript. All authors have read and approved the final version of the manuscript, and agree with the order of presentation of the authors.

Data availability

Data are available from the corresponding author upon reasonable request.

Code availability

Code is available from the corresponding author upon reasonable request.

Competing interests

The authors declare that they have no competing interests.