Physical activity and exercise, the latter a subcategory of physical activity [1], have been recommended as a vital part of standard care throughout the lifespan for those with conditions that include, but are not limited to, inflammatory arthritis and osteoarthritis [2, 3], as well as fibromyalgia [4]. While the overall benefits of physical activity and exercise appear to be well-established overall, the issue of inter-individual response differences (IIRD) to exercise is less clear and has not been appropriately examined. Determining true IIRD is important because it is currently considered to be one of the most important topics in exercise precision medicine [5], an approach that falls under the broader definition of precision medicine, previously defined as “The tailoring of medical treatment to the individual characteristics of each patient…to classify individuals into subpopulations that differ in their susceptibility to a particular disease or their response to a specific treatment” [6].

While previous research has reported IIRD associated with exercise, including genetic interactions, for a variety of outcomes [7], these studies have traditionally placed the “cart before the horse” by not appropriately testing for within-subject random variation first [5]. Put simply, these studies did not determine if the variation in response to exercise was the result of the exercise itself versus random variation (measurement error biological variability, etc.) and/ or behavioral changes (sleep, diet, etc.) not associated with the treatment response to exercise. As a result, this could lead to potentially false conclusions and unethical follow-up studies [8]. From a clinical perspective, this could lead to the recently stated and possibly false suggestion of developing exercise programs based on one’s genotype [7] as opposed to a focus on more general exercise guidelines [2,3,4], the latter of which should have greater reach, especially for marginalized populations.

While different approaches for examining IIRD exist [9], one novel approach is to compare the standard deviations (SDs) of changes between the intervention and control arms from appropriately conducted randomized controlled trials (RCTs), details of which have been described elsewhere [8, 10]. In other words, rather than comparing mean change outcomes (post minus pre) between the exercise and control groups, one compares the changes in the SDs between the exercise and control groups. This means that change outcome SDs are now treated as point estimates.

Using the meta-analytic approach to examine IIRD after performing a traditional meta-analysis, one can take the difference between intervention and control group standard deviations, treated as point estimates, from each study as follows [10]:

$$\sqrt{\mathrm{SD}_{\mathrm i}^2-\mathrm{SD}_{\mathrm c}^2}$$

where \({\text{SD}}_{i}^{2}\) represents the standard deviation for the intervention group and \({SD}_{c}^{2}\) represents the standard deviation for the control group. The standard error of the variance for the SD point estimates are then computed as follows [10]:

$$\mathrm{SE}=\sqrt{2\left(\frac{\mathrm{SD}_{\mathrm i}^4}{{\mathrm{DF}}_{\mathrm i}}+\frac{\mathrm{SD}_{\mathrm c}^4}{{\mathrm{DF}}_{\mathrm c}}\right)}$$

where DF are the degrees of freedom minus 1 for the standard deviations of both the intervention and control groups. Results are then pooled by combining individual response variances and their standard errors into one overall point estimate and 95% confidence interval (CI) using an appropriate pooling model. For both the traditional and IIRD analyses, either the inverse variance heterogeneity (IVhet) or quality effects (QE) models of Doi et al. are recommended, details of which have been described elsewhere [11,12,13,14]. The SD for the pooled point estimates and the 95% CI are then computed by calculating the square root of each [15]. If the lower 95% CI is negative, the sign is initially ignored, the square root computed, and the sign reapplied. Absolute between-study heterogeneity is then calculated using tau (\(\tau\)) [15]. If the SDs are similar or greater in the control group, then IIRD associated with exercise are clinically unimportant, thereby negating any further analysis examining for potential moderators and mediators, including genetic interactions [8].

As a classic example of a lack of true IIRD at the study level, low cardiorespiratory fitness has been shown to be a potent predictor for all-cause mortality as well as coronary heart disease and cardiovascular disease events in healthy men and women [16], with heritability estimates reportedly ranging from 25 to 65% for changes in cardiorespiratory fitness as a result of aerobic exercise [7, 17]. However, a recent review that included 186 published studies found that none had appropriately quantified IIRD, with only one including a control arm comparator [18]. Reanalysis of the one intervention study that included a control arm found that aerobic exercise increased maximum oxygen consumption in ml.kg−1.min−1. However, when true IIRD was appropriately quantified, the standard deviation of change in cardiorespiratory fitness, assessed as maximum oxygen consumption in ml.kg−1.min−1, was found to be greater in the control arm (± 5.6 ml.kg−1.min−1) than the aerobic exercise arm (± 3.7 ml.kg−1.min−1) [18]. Thus, while maximum oxygen consumption improved in the exercise arm, no response variability to exercise, i.e., IIRD, was found. Consequently, the variability observed was the result of random variation (measurement error, biological day-to-day variation) and/or the physiological responses from other factors associated with behavioral changes (sleep, diet, etc.) that were not the result of a treatment such as aerobic exercise [8, 19, 20]. From a research perspective, this suggests that an examination for potential moderators and mediators, including genetic interactions, is probably not warranted. From a clinical viewpoint, these results suggest that more general exercise guidelines may be more appropriate.

While the author is not aware of any previous work by others that has examined for true IIRD as a result of exercise, or any other intervention in rheumatology, our research group recently conducted a meta-analysis to determine if true aerobic exercise-associated IIRD exist with respect to changes in anxiety among adults with fibromyalgia (FM) [21]. Using the approach previously described for determining true IIRD, five RCTs representing a total of 321 adults with FM were included [21]. The IVhet model was used to pool results for both the traditional and IIRD meta-analyses. On a 10-point scale, statistically significant exercise minus control treatment effect reductions were observed for anxiety (\(\overline{X }\), − 0.77 points, 95% CI, − 1.25 to − 0.29, p = 0.02). However, no significant IIRD were observed (\(\overline{X }\), 0.6 points, 95% CI, − 1.2 to 1.5, p = 0.39). In addition, the 95% prediction interval (PI) for true IIRD with respect to what result one might expect in a future study in the same population was − 1.7 to 0.8. Furthermore, based on previously suggested categories [22] and a minimal clinically important difference of 0.45 points, the probability, i.e., percent chance, of a clinically meaningful difference in variability, i.e., IIRD, was 61.5% (only possibly clinically important). Based on our findings, it was concluded that while aerobic exercise is associated with reductions in anxiety among adults with FM, there was a lack of convincing evidence to support the notion that true IIRD exist. From a research perspective, these findings again suggest that a search for potential mediators and moderators associated with aerobic exercise, including genetic interactions, may not be warranted. From a clinical perspective, the use of general exercise guidelines for patients may be preferred with respect to the impact of aerobic exercise on anxiety in adults with FM.

While the previous example of IIRD focused on exercise, it is critically important to understand that this issue extends to any type of intervention, outcome, and disease. Given the former, as well as a dearth of original studies as well as meta-analyses that have examined for true IIRD in rheumatology, it is recommended that future original studies and meta-analyses examine for true IIRD, including meta-analyses of data from previously published meta-analyses. In addition, it is recommended that future RCTs and existing meta-analyses report sample sizes as well as change outcome means and standard deviations for both intervention and control arms so that IIRD meta-analyses can be conducted. For original RCTs examining for IIRD, applied recommendations applicable to any intervention and outcome have been provided by Swinton et al. [23]. Finally, it is the hopes that this brief article will motivate authors, reviewers, editors, and practitioners in rheumatology of both the research and clinical importance of examining for true IIRD.