Correlates and Determinants of Cardiorespiratory Fitness in Adults: a Systematic Review

The review methodology, including the search strategy, data collection, and quality assessment of the included studies, was pre-specified and has been published in a review protocol [108]. The review was registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42017055456). We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines whenever applicable.

Relevant studies were located from different sources: We searched the PubMed, EMBASE, and Cochrane Library databases from inception to Present. The last search was run on 1 February 2017. In addition, we conducted a search for grey literature (Google Scholar). A limited update literature search was performed from 01 February 2017 to 6 June 2019 on PubMed. We used a broad range of search terms for CRF measures (outcome; e.g., “cardiorespiratory fitness”) and general correlates and determinants (exposures; e.g., “health behavior”) to ensure that all potentially relevant articles investigating the factors associated with CRF were included [108]. We also cross-checked the references of the articles selected for full-text screening to locate additional studies. No language, text availability, publication status, or date restrictions were imposed. Eligible indicators of CRF were objectively assessed measures of CRF by submaximal or maximal exercise testing. Therefore, both direct indicators measured via spiroergometry and indirect indicators calculated via metabolic equations of oxygen consumption were included. We only included studies that assessed CRF by treadmill or cycle ergometer. The preferred laboratory measure of CRF (the “gold standard”) is maximal oxygen consumption (VO_2max), which is measured in milliliters per kilogram per minute (ml/kg/min) during exercise and reflects a person’s maximal ability in terms of oxygen uptake, use, and transport [109]. VO_2max is defined as the point when oxygen consumption reaches a plateau and cannot be increased with an increase in effort [83, 110]. However, in exercise testing, such a clear plateau often cannot be achieved, and, instead, the highest obtained VO₂ value, regardless the subject’s effort, (VO_2peak) is used [110]. Because the distinction between these two measures was not always clear in the included studies [111], this review uses the term VO_2max for both VO_2max and VO_2peak.

To structure our search strategy, we used a conceptual framework that locates CRF on the pathway to NCDs and defines and categorizes potential factors associated with CRF. Based on the method recommended by Victoria et al. [112], we proposed this framework by adopting elements of different ecological models [108] (Fig. 1).

As mentioned above, to limit the scope of this systematic review, we focused on individual correlates and determinants of CRF and excluded environmental (e.g., public green spaces) and interpersonal (e.g., social support) factors. The following categories of factors were considered: (1) sociodemographic factors (e.g., age, sex, and education); (2) anthropometric measures (e.g., BMI, weight, and waist circumference [WC]; (3) vital parameters (e.g., resting heart rate [HR] and blood pressure [BP]); (4) comorbidities and medications; (5) biomarkers (e.g., C-reactive protein [CRP]); (6) PA parameters (e.g., leisure-time PA [LTPA]); and (7) other health-related behaviors (e.g., smoking and nutrition). Genetic factors, such as specific genetic variants associated with CRF trainability [113], were not included in this review.

In this review, we included quantitative observational studies (cohort studies, case–control studies, and cross-sectional studies) reporting individual correlates of CRF in the general adult population. The following exclusion criteria were applied: (1) studies on interventions or trials promoting CRF or PA; (2) studies focusing on children (aged 0–11 years), adolescents (aged 12–17 years), or older adults (aged 80+ years), although studies investigating a broad age range were not excluded and may include participants aged younger than 18 years or older than 80 years; (3) studies with highly select groups of participants not representative of CRF/PA in the general population (e.g., specific occupations such as firefighters or special forces, athletes, sports students, or institutionalized populations such as patients in hospitals or nursing homes); (4) studies presenting only univariate results that were unadjusted for confounding; (5) studies that assessed CRF by means other than bicycle ergometer or treadmill; and (6) reviews, letters to the editor, commentaries, and editorials.

We imported all search results into Endnote X7 (Thomas Reuters, USA) reference management software and removed duplicates. At the first stage of screening, we reviewed the titles and abstracts. At the second stage, we assessed full-text articles for eligibility. The assessment was performed by two independent reviewers (NP and KJO at the first stage, JZ and KJO at the second stage). If the reviewers disagreed, a third reviewer was included in the discussion (JDF). Final decisions were made by consensus. From the final included studies, the following information was extracted and transferred to a detailed extraction worksheet, which was developed and pilot tested in advance: study characteristics (title, year, study name, country, and study period), study methods (e.g., study type, sample size, and inclusion/exclusion criteria), study population characteristics (e.g., age range and sex ratio), details of CRF assessment (e.g., test machine and exercise test protocol), analytical methods (e.g., statistical method and type of reported outcome), results, and reported limitations (Additional file 4: Data extraction table). When different multivariable models were reported for a sample, we extracted the fully adjusted models. The data extraction was conducted independently by two authors (JZ and KJO). Disagreements were solved by discussion; if agreement could not be reached, a third author (JDF) was involved, and decisions were made by a simple majority. The limited search update was performed by two authors (JZ and JDF [10% random sample at title and abstract screening, full sample at full text screening]). Information from eligible studies retrieved in the update was extracted to a separate table (Additional file 3: Table OR1).

Additional information was obtained through a request to the author for one study, where further results were mentioned but not presented in detail [49]. The author responded, but this additional information was not used because the results violated the inclusion criteria. We contacted authors of two studies to clarify the direction of a reported association [18] and for further details about the study period [65]. Both of these authors responded, and the additional information was included in this study.

Risk of bias in each study was also assessed independently by two authors (JZ and KJO) using a customized version of the Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies, published by the National Heart, Lung, and Blood Institute at the National Institutes of Health, USA [114]. We categorized risk of bias as “high” when a study reached < 50% of the fulfillment score, “moderate” when a study reached 50–75%, and “low” when > 75% of the criteria were fulfilled. Further details about this risk assessment procedure have been published elsewhere [108].

We compiled summary tables aggregating the existing research on each potential correlate or determinant of CRF. For this purpose, we modified the “semi-quantitative approach” originally proposed by Sallis et al. [115] (adapted by [116, 117]). This approach allowed summary measures to be calculated for each analyzed exposure, even when the heterogeneity among studies was high and no meta-analysis was possible. Following this approach, in the present article, we distinguish between the terms “study” and “sample.” Each article included in this review is referred to as a study. Studies where the results were presented separately for men and women were counted as two samples. Studies that presented results only for the total population (men and women) and those that focused on only men or women were counted as one sample. Hereafter, we use the term “study” to refer to each article included in this review and the term “sample” to refer to each (sex-specific) sample or subsample.

For each sample, a significant direct or inverse association between CRF and the exposure is presented as “+” or “−” in the column “related to CRF”; non-significant associations are shown in the column “unrelated to CRF” (“0”). The findings are summarized by presenting the total number of samples and the numbers of samples with direct, indirect, and non-significant associations. Finally, following an approach applied in other semi-quantitative reviews [115‐117], a summary measure for each association was calculated as follows: Agreement in direction in at least 60% of all samples was graded as a positive (“+”), negative (“−”), or non-significant association (“0”). If none of the categories had a majority of at least 60%, the correlate was assessed as unclear (“?”). Outcomes that were heavily investigated (i.e., in at least ten independent samples) for which there was result agreement in at least 80% of the samples are denoted by “++,” “− −,” or “00.” Summary measures were calculated only if an exposure was investigated in three or more individual samples; otherwise, the correlate was marked as not applicable (“n/a”). Each unique association is reported separately. Separate studies drawing on the same study population and reporting redundant exposures are presented in parentheses and were counted as one unit of analysis. For example, the two studies using the National Health and Nutrition Examination Survey data from the same study period that presented data on the association of age with CRF are both listed in the summary table but were counted as only one sample. In rare cases when separate studies using the same sample had contradictory results, we present the relevant details in a footnote.

We identified a total of 78 articles for inclusion in this review. The complete selection process is shown in the PRISMA Flow Diagram (Fig. 2).

The initial search in PubMed, EMBASE, and the Cochrane Library yielded 3016 records. Following the removal of duplicates, 216 articles were added from Google Scholar and the reference cross-check. After title and abstract screening, 338 records remained for full-text screening. A total of 260 of these records did not fulfil the eligibility criteria. All included studies were published in English. The dates of publication ranged from 1966 to 2017; the vast majority of articles were published after 2000 (see Fig. 3).

General characteristics of the studies reviewed can be found in Table 1.

Most studies were from North America (31) or Europe (29), and there were 14 from Asia and four from Oceania (including Australia). Thus, almost all studies were from high-income countries (74), according to the World Bank classification; only four were from lower-middle or upper-middle-income countries, and none were from low-income countries. Except for one case–control study, all of the studies were cross-sectional (59) or cohort (18) studies. Almost two-thirds of the studies investigated both men and women (48), whereas eight studies reported results only for women and 17 reported results only for men. Five studies did not report the sex ratio of the study participants. The age range across the studies was 13 to 96 years. Sample sizes ranged from 79 to 218,820. A variety of statistical techniques were used to investigate the associations between CRF and potential correlates and determinants, including general linear models such as analysis of variance; analysis of covariance; (multiple) linear, logistic, and quadratic regression; partial correlation; and general estimating equations.

The quality scores assessing the risk of bias in the individual studies ranged from 40 to 100% (Additional file 1: Table OR1.1). Thirty studies were classified as having low risk of bias, 41 were classified as having medium risk of bias, and seven were classified as having high risk of bias.

Half of the studies (39) used directly assessed VO_2max via gas analysis as the outcome variable, 19 reported estimated VO_2max, and 19 used only measures other than VO_2max as the outcome variables. Directly assessed VO_2max via gas analysis was more common in studies with smaller sample sizes. No studies with more than 5000 participants performed gas analysis to asses VO_2max (see Fig. 4).

In addition to VO_2max, the following indicators of CRF were included in the selected studies: maximal physical working capacity (PWC) in Watts, PWC in Watts at variable and fixed HR thresholds (e.g., PWC at 75% of the predicted maximal heart rate or PWC at a heart rate of 170 beats per minute [PWC170]), time in seconds to HR threshold (e.g., exercise test duration to reach an HR of 130 beats per minute), energy expenditure in metabolic equivalents, and total exercise duration (in minutes or seconds). Eleven studies reported results for multiple measures of CRF. Almost two-thirds (48) of the studies applied maximal exercise testing (symptom limited), and one-third (27) applied submaximal testing. The treadmill (44) and the cycle ergometer (34) were commonly used exercise test machines in the included studies. Whereas the majority of European studies utilized cycle ergometers (76%), almost all studies conducted in North America used treadmills for exercise testing (94%; see Fig. 5).

This systematic review aimed to give a detailed overview of the potential individual factors influencing CRF and to analyze the consistency of the results of existing research on this topic. Overall, 3016 records were identified, and 78 articles were ultimately included. We found evidence that CRF decreases with age, is lower among women than among men, and is associated with ethnicity. CRF was positively associated with SES, FEV, and vital capacity, and negatively associated with BMI, weight, WC, body fat, resting HR, systolic BP, and CRP. As expected, CRF was associated with several measures of PA. Furthermore, CRF showed a negative association with smoking and a positive association with alcohol consumption in the majority of the included studies. Age, BMI, WC, PA index, smoking, and education were the most investigated factors (≥ 10 samples). For these factors, results showed no significant sex differences. To our knowledge, this is the first comprehensive review investigating the individual determinants and correlates of CRF. A review conducted by Ortega et al. [98] summarized the relationship between fitness and CVD risk factors among children and adolescents. Another systematic review focused on studies investigating the association between genetic variants and CRF trainability [113]; however, this was beyond the scope of the present review.

To our knowledge, this is the first systematic review investigating the individual correlates and determinants of CRF. The semi-quantitative approach used in this review made it possible to present a concise summary of the results from 78 studies and to provide a comprehensive overview of the current state of research and potential research gaps.

We acknowledge that such a comprehensive review also has limitations. The majority of the studies included in this review were cross-sectional; therefore, statements about causal relationships between exposures and outcomes are limited. Furthermore, not all studies used VO_2max (the “gold standard”) as an outcome, diminishing the comparability of results across the studies. Although other methods for measuring CRF (e.g., exercise test duration) have been shown to be correlated with VO_2max [147, 148], differences in methods could lead to deviating results. It is interesting to note that the included studies that reported the associations between an exposure and various measures of CRF often found similar associations, independent of the measurement method used. Several studies also obtained CRF results using submaximal exercise-based predictive equations. Although it has been found that the results of assessed and predicted CRF can vary [111], a systematic review has demonstrated that these predictive equations are quite accurate [149]. In addition to differences in the method of measurement (indirect vs. direct), different indicators of CRF (e.g., VO_2max vs. others) were not taken into account when discussing the results. This could be considered a potential bias because sensitivity analyses showed significant differences in the association between directly assessed VO_2max, indirectly assessed VO_2max, and other CRF measures for some of the exposures. Variations in the methods used in the statistical analyses may also have affected the results. In this systematic review, we only analyzed the adjusted results from multivariable analyses. Because fewer variables usually reach the defined level of significance in such analyses, compared with univariate analyses [115], a bias toward null findings is possible [150]. Such a bias may be exacerbated by the use of the fully adjusted models.

Because of the wide range of correlates and determinants analyzed, as well as the significant heterogeneity in analytical methods and study samples, it was not feasible to conduct a meta-analysis. Correspondingly, we were unable to show the strength of the associations (effect measures) or to adjust the results for sample size. In addition, variation in adjustment for confounding among the studies was not considered in the interpretation of the results for each outcome, although this difference may have led to divergent results across the studies. Due to restrictions in the search strategy and the intra-study requirements to participate in CRF testing, the study populations usually consist of a relatively healthy population. However, it cannot be ruled out that some of the reported associations are influenced by underlying chronic diseases. Furthermore, the universal perspective of this review did not allow determinants, correlates, mediators, moderators, and confounders to be differentiated [151]. However, the focus of this review was to show the consistency of reported associations. Certain indicators may be underrepresented for specific reasons (e.g., underreporting or limitations of the search strategy). Existing studies have shown that correlates and determinants may differ across the life course [37, 98]. Future studies should investigate whether the results of this systematic review can be confirmed for young people and older adults. Furthermore, although this was beyond the scope of the present review, environmental and interpersonal determinants probably also play an important role in determining or moderating CRF and should be further investigated [103].