Association of physical activity with loss of knee joint space width over two years: a compositional data analysis in the Osteoarthritis Initiative

SUMMARY

Objective:

There is continued debate as to how engaging in physical activity (PA), including moderate-to-vigorous PA (MVPA), light PA (LPA), and sedentary time (SED), affects one’s risk for knee osteoarthritis (OA). Traditional regression methods do not account for the codependence of these categories of PA, whereby when one category increases, the others must decrease. Thus, we used compositional data analysis (CoDA) to examine time spent in each category of PA, or PA composition, and its association with loss of knee joint space width (JSW), a common indicator of knee OA progression.

Methods:

We performed a secondary analysis of data from a subset of participants in the Osteoarthritis Initiative. These participants had minute-by-minute activity data collected over 7 days at baseline; we then categorized each minute as MVPA, LPA, or SED. Our exposure, PA composition, represented min/day spent in each category. Our outcome, medial JSW loss, was the difference in medial tibiofemoral JSW from baseline to 2 years later. We employed CoDA, using an isometric log-ratio transformation, to examine the association of PA composition with medial JSW loss over 2 years, adjusting for potential confounders.

Results:

We included 969 participants (age: 64.5 years, 56% female, body mass index [BMI]: 28.8 kg/m²). Mean PA composition was: MVPA 9.1 min/day, LPA 278 min/day, SED 690 min/day. Per adjusted regression models, higher MVPA was not associated with greater medial JSW loss (β = −0.0005, P = 0.97), nor was LPA (β = 0.06, P = 0.27) or SED (β = −0.06, P = 0.21).

Conclusion:

Using CoDA, PA composition was not associated with medial JSW loss over 2 years.

Introduction

Knee osteoarthritis (OA) is often considered to be a mechanically driven disease. As such, reduced loading of the knees has been considered a possible prevention strategy, which could include limiting daily physical activity (PA). Early studies concluded that higher engagement in PA increased one’s risk for structural progression of knee OA^1–3. However, more recent studies have highly mixed findings; they concluded that PA could be harmful^4,5, beneficial^6–8, or have no effect^9–12 on risk for knee OA progression.

Why did these studies, many of which used the same large data sets, yield such mixed findings? One possible explanation relates to how they defined, analyzed, and/or categorized PA. Generally, PA is defined as “any bodily movement produced by skeletal muscle that results in energy expenditure”¹³, and is often divided into categories based on the intensity of movement: 1) moderate-to-vigorous intensity PA (MVPA), which includes moderate activities like brisk walking and jogging, and more vigorous activities like running; 2) light intensity PA (LPA), which includes slow walking and light housework; and 3) sedentary time (SED), which includes quiet standing, sitting, and reclining¹⁴.

Previous studies focused on one or two categories of PA (e.g., MVPA, SED) and examined how they independently affected the knee joint; that is, an assumption was made that one category (MVPA, LPA, and SED) could vary independently of the other categories. However, if a person were to increase time in MVPA, they would also decrease time in LPA and/or SED because the total time in a day is fixed (1,440 min). Thus, to truly understand the effect of higher MVPA on the knees, one must also account for the effects of lower LPA and/or SED. By analyzing the “complete picture” of PA, one can avoid potentially misleading results that occur when categories of PA are examined independently^15–17.

Thus, we aimed to evaluate the association of PA composition, which accounts for time in MVPA, LPA, and SED, with loss of knee joint space width (JSW), an indicator of structural progression of knee OA. Specifically, we examined how a simulated difference in a category of PA (and corresponding difference in the other categories) was associated with a difference in JSW loss over the following 2 years. First, we hypothesized that spending more time in MVPA (and less time in LPA and SED) would not be associated with JSW loss over 2 years. This is because previous studies that used our same data set (from the Osteoarthritis Initiative), though they did not use compositional data analysis (CoDA), most consistently found no independent association of MVPA with JSW loss^10–12.

Second, we hypothesized that spending more time in LPA (and less time in MVPA and SED) would be associated with lower JSW loss. Previous studies have not examined the independent association of LPA with JSW loss, but it has been suggested that low-impact activities (which often fall in the category of LPA) promote cartilage health¹⁸. Third, we hypothesized that spending more time in SED (and less time in MVPA and LPA) would be associated with greater JSW loss. We suspected that unloading the knees during SED, coupled with minimal loading throughout the rest of the day (i.e., less time in MVPA and LPA), could lead to early degenerative changes in knee cartilage^4,19 and/or cartilage thinning^20–22.

Methods

Study design

We performed a secondary analysis of longitudinal data from the Osteoarthritis Initiative (OAI), an NIH/NIA-funded cohort study of 4,796 adults with (or at risk for) knee OA at enrollment²³. OAI data are publicly available at https://nda.nih.gov/oai/.

Setting

Adults enrolled in the OAI between 2004 and 2006 at study sites in Columbus, Ohio, Baltimore, Maryland, Pittsburgh, Pennsylvania, and Pawtucket, Rhode Island. All clinical data and radiographic images were collected at these sites. Institutional review boards at all sites approved the study protocol, and all participants provided written informed consent²³.

Participants

Details about recruitment and inclusion/exclusion criteria for the OAI can be found in its study protocol²³. Briefly, the OAI enrolled adults aged 45–79 years, who 1) had a radiographic diagnosis and symptoms of knee OA (Progression cohort), or 2) had major risk factors but no diagnosis or symptoms of knee OA (Incidence cohort), or 3) had no major risk factors and no diagnosis or symptoms of knee OA (Reference [Nonexposed] cohort)²³.

A subset of 2,127 OAI participants, including adults from all three cohorts, enrolled in an ancillary study²⁴ in which investigators collected their minute-by-minute accelerometry data at the 48-month visit. Because this was the first visit at which accelerometry data were collected, we refer to the 48-month visit as the study baseline.

Key variables

Exposure: PA composition at baseline

To determine the PA composition of each participant, we assessed time (min/day) at baseline in each of the following, mutually exclusive categories: 1) MVPA, 2) LPA, and 3) SED. PA was assessed for 7 days using the ActiGraph GT1M monitor (ActiGraph LLC, Pensacola, FL), a hip-worn accelerometer that produces a weighted sum of activity counts each minute. An activity count is a unit of acceleration (in the vertical plane) that correlates to the speed or intensity of bodily movement, thus allowing us to categorize each minute as MVPA, LPA, or SED. The Actigraph GT1M monitor provides a valid²⁵ and reliable²⁶ assessment of PA in adults, including those with knee OA²⁷.

First, we applied a definition of nonwear (i.e., time that we assumed the participant was not wearing the monitor) that was developed using OAI data²⁷: a period of at least 90 consecutive minutes with zero activity counts, allowing for up to 2 nonconsecutive minutes with low activity counts (<100 counts/min). After removing periods of nonwear, we restricted our analysis to participants who wore the ActiGraph GT1M monitor for 4–7 days²⁸ and at least 10 h/day²⁹, as this is needed to provide a valid representation of their PA.

Next, we categorized each minute of wear time in a day using established cut-points²⁸: 1) MVPA (≥2020 counts/min), 2) LPA (100–2019 counts/min), and SED (0–99 counts/min). We then divided the minutes in each category (MVPA, LPA, and SED) by the total wear time for that day (e.g., 10 h = 600 min). Doing so created a unitless proportion, or component, of wear time for that day; collectively, we refer to these components as PA composition. We determined PA composition (including MVPA, LPA, and SED) for each participant, then used these data to perform CoDA.

Outcome: medial JSW loss from baseline to two years

Over 2,000 OAI participants had radiographs collected and assessed at our study baseline (48-month visit) and 2 years later (72-month visit). In contrast, a much smaller subset of OAI participants had longitudinal magnetic resonance (MR) images assessed and made publicly available through ancillary studies. Therefore, we leveraged the availability of radiographs in the OAI cohort and used JSW in the central medial tibiofemoral (TF) joint, at fixed location χ = 0.25, as a surrogate measure for cartilage morphology (typically measured via MR images). JSW at χ = 0.25 demonstrates the highest reliability (intra-class correlation coefficient [ICC] > 0.98), is most sensitive to change across all grades of OA severity³⁰, and is most likely to represent cartilage thinning³¹ of the nine JSW locations.

We estimated medial JSW loss (mm) by subtracting JSW at 2 years (mm) from JSW at baseline (mm). JSW loss is considered a valid, quantitative measure of OA progression³². To ensure that only high-quality radiographs were used to estimate medial JSW loss (mm), we excluded radiographs with an improper beam angle (<5° or >15°, variable V0XBMANG) or excessive tibial rim distance (>6.5 mm, variable V0XTPCFDS) at either time point³³. We also excluded radiographs with an excessive difference in beam angle (>5°) or tibial rim distance (>1.4 mm) between the two time points³³. If radiographs for both knees were considered high-quality, the right knee was randomly selected for inclusion in our analysis (hereafter referred to as the study knee).

Additional inclusion/exclusion criteria

At baseline, we excluded OAI participants for any of the following reasons: 1) they enrolled in the OAI accelerometry study but did not wear and return the Actigraph monitor; 2) they wore the Actigraph monitor, but did not have valid PA data (i.e., they had only 1—3 days of wear time); 3) baseline JSW was missing; 4) baseline JSW was not valid due to the presence of a prosthesis, poor image quality, improper beam angle, and/or excessive tibial rim distance; or 5) baseline JSW was 0 mm (Fig. 1).

Fig. 1.

Flow diagram of OAI participants who were included in our analyses.

At 2 years, we excluded OAI participants who did not have valid medial JSW loss because 1) 2-year JSW was missing, 2) 2-year JSW was not valid due to poor image quality, improper beam angle, excessive tibial rim distance, or 3) medial JSW loss was not valid due to excessive variation in beam angle or tibial rim distance from baseline to 2 years (Fig. 1).

Potential confounders

Our primary analysis was adjusted for the following potential confounders, as they could be associated with PA and medial JSW loss in the study knee: age (years)³⁴, sex (male/female)³⁵, body mass index (BMI, kg/m²)³⁴, race (White/Black/Asian/Non-white)³⁵, previous injury that limited walking for at least 2 days (yes/no)³⁶, previous surgery (yes/no)³⁷, baseline medial JSW (mm)³⁸, and lateral JSW loss (mm).

Potential sources of bias

Participants with missing exposure data (PA composition) and/or outcome data (medial JSW loss over 2 years) were excluded from our analyses. To evaluate the potential for selection bias (due to the exclusion of participants with missing data), we described the characteristics of adults who were included vs excluded from our analyses (Supplementary Table 1).

Statistical analysis

We examined the association of PA composition (min/day in MVPA, LPA, and SED) with medial JSW loss (mm) over 2 years using CoDA (see Supplementary File for details). All analyses were performed using SAS Analytics Software 9.4.

Briefly, we used an established isometric log-ratio (ilr) approach to transform each PA component (MVPA, LPA, and SED) into a set of two ilr coordinates³⁹. We entered each set of coordinates as explanatory variables in a multiple linear regression model, adjusting for potential confounders (age, sex, BMI, race, previous knee injury or surgery, baseline medial JSW, and lateral JSW loss).

Thus, we generated a total of three separate compositional regression models in our primary analysis. Model 1 simulated both increasing time in MVPA at intervals of 2 min/day and decreasing the total time spent in LPA and SED at the same intervals (2 min/day). The model did not assume that time in LPA and SED decreased equally (e.g., −1 min/day of LPA, −1 min/day of SED); rather, it used the existing ratio of LPA to SED (e.g., 1:4), and decreased time accordingly (e.g., −0.5 min/day of LPA, −1.5 min/day of SED). The model increased time in MVPA at intervals of 2 min/day until the maximum value of MVPA (in min/day) for the sample was reached. Similarly, the model simulated decreasing time in MVPA at intervals of 2 min/day until the minimum value of MVPA for the sample was reached. Using the full range of values for MVPA from the sample, the model estimated the association of simulated differences in PA composition (i.e., time in MVPA, LPA, and SED) with a difference in medial JSW loss over 2 years.

We repeated the process described above, but simulated increasing (or decreasing) time in LPA (Model 2) and SED (Model 3) at intervals of 10 min/day.

Each model was evaluated using an isotemporal plot (Figs. 2–4). For each plot, the origin for the x-axis (x = 0) corresponded to the mean PA composition for the sample (i.e., min/day in MVPA, LPA, and SED). The origin for the y-axis (y = 0) corresponded to the mean medial JSW loss (mm) for the sample. Thus, each plot depicted differences in a specific PA category (in min/day) relative to the sample’s mean PA composition, then estimated the effect on medial JSW loss relative to the sample’s mean medial JSW loss.

Fig. 2.

Isotemporal plot for Model 1, simulating a difference in MVPA (on the x-axis, in min/day) and its association with a difference in medial JSW loss (on the y-axis, in mm). The origin (0,0) corresponds to the sample mean composition (MVPA: 9.1, LPA: 278, SED: 690 min/day), and sample mean JSW loss (0.13 ± 0.5 mm). The range of x-values (−9.1 to +30 min/day) indicates a difference in MVPA relative to the sample mean composition. Negative y-values indicate lower medial JSW loss, or a “better” outcome.

Fig. 4.

Isotemporal plot for Model 3, simulating a difference in SED (on the x-axis, in min/day) and its association with a difference in medial JSW loss (on the y-axis, in mm). The origin (0,0) corresponds to the sample mean composition (SED: 690, MVPA: 9.1, LPA: 278 min/day), and sample mean JSW loss (0.13 ± 0.5 mm). The range of x-values (−120 to +120 min/day) indicates a difference in SED relative to the sample mean composition. Negative y-values indicate lower medial JSW loss, or a “better” outcome.

We limited the range of values for the x-axis of each isotemporal plot, as follows: MVPA: ±30 min, LPA: ±120 min, SED: ±120 min. This was done to illustrate feasible, yet potentially meaningful, increases or decreases in that specific PA category. However, as mentioned above, the regression coefficients were generated using the complete ranges of values of MVPA, LPA, and SED (min/day) for the sample.

Sensitivity analysis

We separately examined the association of PA composition with lateral JSW loss (fixed location x = 0.75) over 2 years. We applied the same exclusion criteria to radiographs as the primary analysis and adjusted for the same potential confounders.

Results

Sample

We included 969 OAI participants (mean age: 64.5 years, 56% female, mean BMI: 28.8 kg/m², 83% white) who had valid PA data at baseline and valid medial JSW loss over 2 years (Fig. 1). Characteristics of our sample can be found in Table I. Generally, those included in our analyses had similar characteristics to those who were excluded (Supplementary Table 1).

Table I.

Characteristics of OAI participants included in our analyses (n = 969). Unless indicated otherwise, values represent the mean ± SD or n (%)

Characteristic	Value
Age (years)	64.5 ± 8.8
Sex (female)	542 (56)
BMI (kg/m2)	28.8 ± 4.9
Race (white)	803 (83)
Previous knee injury (yes, study knee)	320 (33)
Previous knee surgery (yes, study knee)	148 (15)
OAI cohort assignment
Progression	298 (31)
Incidence	666 (68)
Reference (Nonexposed)	5 (1)
PA composition
MVPA (minutes/day, geometric mean (range))	9.1 (0-130)
LPA (minutes/day, geometric mean (range))	278 (39-647)
SED (minutes/day, geometric mean (range))	690 (222-1017)
Total wear time (minutes/day, geometric mean (range))	977 (635-1365)
Medial JSW at baseline (mm)	5.3 ± 1.2
Medial JSW loss over two years (mm)	0.13 ± 0.5
Percent medial JSW loss over two years (%)	2.6 ± 10

Primary analysis: PA composition and medial JSW loss (mm)

The mean PA composition for the sample (Table I) indicates that on average, participants spent their wear time as follows: MVPA = 0.9%, LPA = 28%, and SED = 71%, or, in decimal form: MVPA = 0.009, LPA = 0.28, SED = 0.71.

Each model examined a simulated difference in a specific PA category (e.g., MVPA) and the corresponding, simulated differences in the other categories (e.g., LPA and SED), to determine their effect on medial JSW loss over 2 years (mm). In Table II, we report the adjusted regression coefficient (β) for the first ilr coordinate of each model (see Supplementary File for details), which indicates the magnitude and direction of the association.

Table II.

Adjusted regression coefficients, 95% CI limits, and P-values for the first ilr coordinate of each compositional regression model in the primary analysis. To interpret each regression coefficient (β), we used back transformation to calculate the expected difference in medial JSW loss (mm), given a simulated increase in a specific PA category (with corresponding decreases in the other categories)

		Expected Difference in Loss of JSW, mm (95% CI), Given a Simulated Increase of:
Regression Model	β (95% CI) p-value	+30 min	+60 min	+90 min	+120 min
Model 1. MVPA (with corresponding differences in LPA and SED)	−0.0005 (−0.03, 0.03) p = 0.97	−0.0007 (−0.04, 0.04)	−0.0009 (−0.05, 0.05)	−0.001 (−0.06, 0.06)	−0.001 (−0.07, 0.07)
Model 2. LPA (with corresponding differences in SED and MVPA)	0.06 (−0.05, 0.2) p = 0.27	0.008 (−0.006, 0.02)	0.015 (−0.01, 0.05)	0.02 (−0.02, 0.07)	0.03 (−0.02, 0.09)
Model 3. SED (with corresponding differences in MVPA and LPA)	−0.06 (−0.2, 0.04) p = 0.21	−0.008 (−0.03, 0.005)	−0.02 (−0.05, 0.01)	−0.03 (−0.08, 0.02)	−0.04 (−0.1, 0.02)

Table II also shows the interpretation of each regression coefficient following back transformation; that is, we report the expected difference in medial JSW loss (mm), given an increase in MVPA (Model 1), LPA (Model 2), or SED (Model 3) of 30, 60, 90, or 120 min/day, with corresponding differences in the other categories.

Per Model 1, a simulated increase in MVPA relative to the sample mean (9.1 min/day), with corresponding decreases in LPA and SED, was not associated with a difference in medial JSW loss over 2 years (β = −0.0005, P = 0.97).

Per Model 2, a simulated increase in LPA relative to the sample mean (278 min/day), with corresponding decreases in MVPA and SED, was associated with minimally greater medial JSW loss over 2 years (β = 0.06, P = 0.27), though this did not reach statistical significance. That is, the positive regression coefficient for Model 2 indicated that increased time in LPA was associated with greater medial JSW loss, or a “worse” outcome.

Per Model 3, a simulated increase in SED relative to the sample mean (690 min/day), with corresponding decreases in MVPA and LPA, was associated with minimally lower medial JSW loss over 2 years (β = −0.06, P = 0.21), though this also did not reach statistical significance. That is, the negative regression coefficient for Model 3 indicated that increased time in SED was associated with lower medial JSW loss, or a “better” outcome.

Sensitivity analysis: PA composition and lateral JSW loss (mm)

We generated three additional regression models, replacing medial JSW loss (mm) with lateral JSW loss (mm) as our outcome. We again evaluated each model using the regression coefficient (β) for the first ilr coordinate, and found little to no association between PA composition and lateral JSW loss over 2 years (Supplementary Table 2).

Isotemporal plots

Using back transformation, we created isotemporal plots for each compositional regression model in the primary analysis (Figs. 2–4). Each plot depict a simulated difference in PA composition (on the x-axis) relative to the sample mean PA composition, and the expected difference in medial JSW loss (on the y-axis, in mm) relative to the sample mean JSW loss.

Discussion

In this study, we address an important question: how does PA affect the structural integrity of the knees? This question remains both a topic of debate among researchers⁴⁰ and a major source of uncertainty for healthcare providers and patients⁴¹. Our objective was to comprehensively evaluate PA by accounting for time in all PA categories, i.e., PA composition, since more time in one category (e.g., MVPA) necessarily means less time in others (e.g., LPA and SED).

Overall, PA composition had little to no association with either medial or lateral JSW loss over 2 years, though we will focus our discussion on the findings of the primary analysis (medial JSW loss). First, we found that a simulated increase in MVPA (+120 min/day) beyond the sample mean (9.1 min/day), with corresponding reductions of LPA (−34 min/day) and SED (−86 min/day), was not associated with any appreciable difference in medial JSW loss (−0.001 ± 0.07 mm) over 2 years. Thus, our first hypothesis was supported, as CoDA demonstrated that higher MVPA was not associated with medial JSW loss. Further, our finding is consistent with those of recent observational studies that used traditional regression methods^9–12.

Second, we found that a simulated increase in LPA (+120 min/day) beyond the sample mean (278 min/day), with reductions of MVPA (−1.5 min/day) and SED (−118.5 min/day), was associated with minimally greater medial JSW loss (0.03 ± 0.05 mm) over 2 years. That is, adults with higher LPA were expected to lose 0.16 mm of medial JSW over 2 years, as compared with a mean loss of 0.13 mm for the sample. Of note, the standard deviation for the sample was 0.5 mm, and the limits of agreement for medial JSW loss in the OAI (over one to 3 years) are +0.550 mm and −0.637 mm⁴². Thus, this small magnitude of effect on medial JSW loss has questionable clinical relevance, in addition to not being statistically significant. Further, our second hypothesis, that higher LPA would be associated with lower medial JSW loss, was not supported. Previous studies had not examined the effects of LPA on structural progression of knee OA, but rather focused on MVPA and SED. Therefore, our finding that higher LPA had minimal to no effect on medial JSW loss, made possible using CoDA, is highly novel for this question.

Third, we found that a simulated increase in SED (+120 min/day) beyond the sample mean (690 min/day), with reductions of MVPA (−4 min/day) and LPA (−116 min/day), was associated with minimally lower medial JSW loss (0.04 ± 0.06 mm) over 2 years. Like our finding for LPA, such a small magnitude of effect on medial JSW loss has questionable clinical relevance, in addition to not being statistically significant. Thus, our third hypothesis, that higher SED would be associated with greater medial JSW loss, was not supported. While this finding was generated using CoDA, a similar finding was made in a recent observational study in the OAI that used traditional regression methods to examine SED and worsening KL grade on radiographs¹¹.

Importantly, our findings indicate that adults with knee OA could follow the 2018 PA guidelines from the United States Department of Health and Human Services (DHHS) without raising concerns about accelerating structural OA progression. These guidelines recommend that all adults engage in at least 150 min/week of MVPA, or an average of 22 min/day¹⁴. Unfortunately, most adults fail to meet this recommendation, which was true of our sample; 14% engaged in 0 min/day of MVPA and 69% engaged in less than 22 min/day of MVPA. However, our findings indicate that increasing MVPA up to, and beyond, 22 min/day (with consequently less LPA and SED) was not associated with greater medial JSW loss, even when accounting for baseline JSW as a potential confounder. Therefore, adults with knee OA should be encouraged to engage in MVPA; based upon these findings, MVPA does not seem to harm the structural integrity of the knees, and conversely, has well-established benefits in terms of OA-related pain and function⁴⁰.

The 2018 PA guidelines also recommend that all adults engage in LPA, although there is no time-based recommendation (e.g., min/week). LPA may be more attainable for some adults, and it still offers some functional benefits⁴³. Our findings indicated that like MVPA, higher LPA had little to no association with medial JSW loss and thus, may be safe for adults with knee OA. However, unlike MVPA, LPA may not be associated with improvements knee pain⁴⁴, nor does it offer the same general health benefits⁴⁵. That is, while LPA does not appear to be harmful to the structural integrity of the knees based on our findings from CoDA, it may have less overall benefit for adults with knee OA as compared with MVPA.

Finally, we want to acknowledge some limitations of the current study. First, OAI participants were instructed to wear the ActiGraph GT1M monitor during waking hours and remove it while they slept. Therefore, we could not account for all 1,440 min of each day, and it is possible that participants engaged in MVPA, LPA, or SED after they removed the monitor. Second, as with nearly all activity monitors, the ActiGraph GT1M monitor does not distinguish between sitting and quiet standing⁴⁶, so both are categorized as SED. However, sitting and standing place different magnitudes of load on the knees, so future studies should consider using a monitor that distinguishes between sitting and standing⁴⁷.

Third, we acknowledge that JSW loss does not solely reflect cartilage degradation, which is ideally measured using MR images, nor does it reflect clinical progression (i.e., worsening symptoms) of knee OA. However, few participants had MR images available, and studies show that medial JSW loss likely represents cartilage thinning and radiographic OA progression³¹, even if not clinical progression. Fourth, we restricted our follow-up period to 2 years because we felt it most appropriate to examine the short-term effects of PA composition (measured at baseline) since PA habits can change over time (observed in 20% of OAI participants⁴⁸). Our sample still experienced a mean medial JSW loss of 0.13 ± 0.5 mm over 2 years, which was comparable to that of a previous study (0.14 ± 0.78 mm)⁴⁹.

Finally, there is some potential for selection bias in this study; for example, we had to exclude ~25% of OAI participants with valid PA data at baseline because they were missing baseline JSW. However, those who were excluded based on the availability and/or quality of their radiographs at baseline or 2 years were similar to those who were ultimately included in our analyses.

Despite these limitations, our study has some notable strengths. First, we used an analytical technique (CoDA) that comprehensively evaluated all categories of PA (i.e., PA composition). Previous studies used traditional regression methods to examine independent associations of one or two categories of PA, and could not account for increases in one category and compensatory decreases in other categories. Since the emergence of CoDA in the field of PA, researchers have expressed concern that applying traditional regression methods to inherently compositional data can yield inconsistent or misleading results^15–17. Thus, we leveraged CoDA to account for the compositional nature of PA data, or put another way, the codependence of PA categories.

The second strength of this study relates to generalizability, as the mean PA composition for our sample was highly representative of older adults⁵⁰ and the distribution of PA composition was wide; MVPA ranged from 0 to 130 min/day, LPA ranged from 39 to 647 min/day (~1 to 11 h/day), and SED ranged from 222 to 1,017 min/day (~4 to 17 h/day). Our findings are generalizable to the many adults, with or at risk for knee OA, who have similar characteristics to those enrolled in the OAI.

The final strength of this study is our use of data from the OAI, a large, well-established cohort study with robust standardized procedures for assessment of PA and JSW loss.

In summary, a simulated increase in MVPA (with reductions in LPA and SED) was not associated with a difference in medial JSW loss over 2 years. Similarly, increases in LPA or SED had minimal effects on medial JSW loss over 2 years, as the magnitudes of difference (in mm) were very small. Thus, we recommend that adults, particularly those with or at risk for knee OA, strive to engage in MVPA, as doing so does not seem to affect structural progression of knee OA.

Fig. 3.

Isotemporal plot for Model 2, simulating a difference in LPA (on the x-axis, in min/day) and its association with a difference in medial JSW loss (on the y-axis, in mm). The origin (0,0) corresponds to the sample mean composition (LPA: 278, SED: 690, MVPA: 9.1 min/day), and sample mean JSW loss (0.13 ± 0.5 mm). The range of x-values (−120 to +120 min/day) indicates a difference in LPA relative to the sample mean composition. Positive y-values indicate greater medial JSW loss, or a “worse” outcome.