Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2023 Jan 5;75(6):1340–1350. doi: 10.1002/acr.25021

Association Between Change in Body Mass Index and Knee and Hip Replacements: A Survival Analysis of Seven to Ten Years Using Multicohort Data

Zubeyir Salis 1,✉, Amanda Sainsbury 2
PMCID: PMC10953021  PMID: 36106942

Abstract

Objective

To define the association between change in body mass index (BMI) and the risk of knee and hip replacement.

Methods

We used data from 3 independent cohort studies: the Osteoarthritis Initiative (OAI), the Multicenter Osteoarthritis Study (MOST), and the Cohort Hip and Cohort Knee (CHECK) study, which collected data from adults (45–79 years of age) with or at risk of clinically significant knee osteoarthritis. We conducted Cox proportional hazards regression analysis with clustering of both knees and hips per person to determine the association between change in BMI (our exposure of interest) and the incidence of primary knee and hip replacement over 7–10 years’ follow‐up. Change in BMI (in kg/m2) was calculated between baseline and the last follow‐up visit before knee or hip replacement, or for knees and hips that were not replaced, the last follow‐up visit.

Results

A total of 16,362 knees from 8,181 participants, and 16,406 hips from 8,203 participants, were eligible for inclusion in our knee and hip analyses, respectively. Change in BMI was positively associated with the risk of knee replacement (adjusted hazard ratio [HRadj] 1.03 [95% confidence interval (95% CI) 1.00–1.06]) but not hip replacement (HRadj 1.00 [95% CI 0.95–1.04]). The association between change in BMI and knee replacement was independent of participants’ BMI category at baseline (i.e., normal, overweight, or obese).

Conclusion

Public health strategies incorporating weight loss interventions could reduce the burden of knee but not hip replacement surgery.

INTRODUCTION

The disease burden of osteoarthritis is significant. It is estimated that 1.0–2.5% of the gross national product of countries such as the US, Canada, the UK, France, and Australia is spent to cover the health costs of osteoarthritis, even without including indirect costs such as absenteeism or early retirement (1, 2). The largest part of this health cost of osteoarthritis is knee and hip replacements (3, 4). Knee and hip replacements are common in individuals with overweight or obesity (5). For the knee, a meta‐analysis found that compared to individuals of normal weight, those with overweight or obesity have a 2.5‐ and 5.5‐fold greater risk of requiring knee replacement, respectively (6). For the hip, a prospective cohort study found that compared to individuals of normal weight, those with overweight or obesity have a 1.4‐ to 1.6‐ and 2.1‐ to 3.3‐fold greater risk of requiring hip replacement, respectively (7).

SIGNIFICANCE & INNOVATIONS.

  • Decrease in body mass index (BMI) is associated with reduced risk of knee but not hip replacement.

  • Increase in BMI is associated with increased risk of knee but not hip replacement.

  • Public health strategies incorporating weight loss interventions could reduce the burden of knee but not hip replacement surgery.

While overweight and obesity are clear risk factors for knee and hip replacement, it is not clear whether weight loss reduces the incidence of knee and hip replacement. There are 3 research studies that have investigated this to date (8, 9, 10), each using a prospective cohort, and the results are mixed. The first study used data from the Osteoarthritis Initiative (OAI) cohort and found an association between weight loss and reduced risk of both knee and hip replacement (8). The second study used data from an Australian population‐based cohort and found an association between weight loss and reduced risk of knee but not hip replacement (9). The third study also used data from the OAI cohort but found no association between weight loss and knee or hip replacement (10).

All of the 3 cohort studies mentioned above used data from a single, prospective cohort. Therefore, to clarify and extend the findings from those 3 studies, we combined data sets from 3 prospective cohort studies: the OAI study and the Multicenter Osteoarthritis Study (MOST) from the US, as well as the Cohort Hip and Cohort Knee (CHECK) study from the Netherlands. Change in BMI was our exposure of interest rather than change in weight because all 3 cohorts had BMI data but not weight data.

MATERIALS AND METHODS

Data

We obtained publicly available data from the OAI (11) as well as the MOST study (12) from the US and the CHECK study (13) from the Netherlands. All 3 of these cohorts consisted mainly of participants with or at risk of knee osteoarthritis. Additionally, CHECK had some participants with or at risk of hip osteoarthritis. There was no statistically significant heterogeneity between the 3 cohorts in the effect of our exposure of interest (change in BMI, as detailed below) on our outcomes of interest (knee replacement and hip replacement, as detailed below, and see meta‐analysis in Supplementary Appendix A, available on the Arthritis Care & Research website at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021). The definitions of the exposure and outcomes used for this study did not differ between the 3 study cohorts. The follow‐up times for each of the 3 cohorts ranged from 7–10 years (8 years for OAI, 7 years for MOST, and 10 years for CHECK). The covariates that we selected for adjustment of the analyses were defined similarly in the OAI and MOST study, with some differences in the CHECK study. However, we were able to map the values of variables used in the CHECK study to the values of variables used in the OAI and MOST study. Supplementary Appendix A, available at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021, provides detailed information about each of these 3 cohorts and the covariates used for our analyses. The informed consent documents as well as ethics approvals were reviewed and approved by the local institutional review boards of the studies mentioned above.

Participants and study design

Figure 1 shows the selection criteria for this study. Of the 8,824 participants from the 3 cohort studies mentioned above, participants who either had cancer at baseline or who developed cancer during follow‐up were excluded to avoid any influence of pathologic weight loss. Similarly, participants with a BMI <18.5 kg/m2 at any time (including baseline) during the 7–10 years’ follow‐up (8 years for OAI, 7 years for MOST, and 10 years for CHECK) were also excluded, as this may indicate pathologic weight loss. Participants with no baseline BMI were also excluded, as the change in BMI to 7–10 years’ follow‐up could not be determined for those participants. After these participant‐level exclusion criteria were applied, the remaining 8,318 participants were then placed into knee and hip cohorts. There were 16,636 knees and 16,636 hips from these 8,318 participants in the knee and hip cohorts, respectively. Of these 8,318 participants in each cohort, the participants who had any knee replacement prior to baseline were excluded from the knee cohort, and those who had any hip replacement prior to baseline were excluded from the hip cohort. This was true even if only 1 knee or 1 hip had been replaced to avoid the possible confounding effect on the remaining (intact) knee or hip, respectively, due to possible gait alterations from having 1 knee or hip replaced (see below for the details of the sensitivity analyses).

Figure 1.

Figure 1

Open in a new tab

Selection of participants and knees and hips for analyses. Separate analyses were conducted for knees and hips. BMI = body mass index; CHECK = Cohort Hip and Cohort Knee (study); HR = hip replacement; KR = knee replacement; MOST = Multicenter Osteoarthritis Study; OAI = Osteoarthritis Initiative (study).

Exposure

Our exposure of interest was change in BMI (decrease or increase, in kg/m2) from baseline to the last follow‐up visit before knee or hip replacement, or for knees and hips that were not replaced, the last follow‐up visit. We used change in BMI instead of change in weight because weight data were not available from all 3 cohorts. We calculated BMI based on the weight and height of the participants, as measured by staff at the study centers for the 3 cohort studies underpinning this study when the participant visited the study center (i.e., at the baseline visit and at the follow‐up visits).

Change in BMI from baseline to the time of any knee or hip replacement was assumed to be that which was calculated from the last follow‐up visit before replacement. The exception to this rule was when the last follow‐up visit before replacement was within 6 months prior to replacement, in which case BMI from the second‐to‐last follow‐up visit was used to calculate change in BMI from baseline. This is due to the common practice of asking patients who are scheduled to have knee or hip replacement to lose weight prior to surgery (14, 15, 16), potentially introducing information bias (17). Additionally, knees or hips that were replaced in the first 18 months from baseline were effectively removed from the analysis (i.e., censored). This is because the first follow‐up visit was at 12 months, and the additional 6 months’ waiting period prior to surgery, as explained above, meant that change in BMI from baseline could only be measured for joint replacements that happened after 18 months.

If BMI data were missing at the relevant follow‐up visit before replacement, then data for the change in BMI from baseline to replacement were classified as missing, but the joint was retained in the analysis. Sensitivity analyses for missing data were performed as detailed below.

Outcome

The outcome event for the survival analyses in this study was the incidence of primary knee or hip replacement from any cause (i.e., all‐cause, not specific to replacements related to osteoarthritis) because this information was not available from all 3 cohorts. For the same reason, both total and partial primary knee and hip replacements were included (without distinction between them) in the analyses. We only included knee or hip replacements that occurred during the 7–10 years of follow‐up.

Population attributable fraction (PAF)

In addition to investigating knee and hip replacements, we estimated the PAF (18), specifically, the proportion of knee and hip replacements that would have been avoided if all knees and hips in that population has been exposed to a particular decrease in BMI. For these estimations, we selected a decrease of 1 BMI unit (kg/m2) from baseline to follow‐up at 7–10 years. We performed the estimations for individuals with overweight and obesity (25 kg/m2 ≤BMI), for individuals with a normal BMI (18.5 kg/m2 ≤BMI <25.0 kg/m2), and for individuals with or above normal BMI (18.5 kg/m2 ≤BMI).

Statistical analysis

Semiparametric time‐to‐event survival analyses, namely Cox proportional hazards models, were used to determine the association between change in BMI and knee and hip replacement. Univariate (unadjusted) and multivariable (adjusted) analyses were performed. For the multivariable analyses, we selected variables using the method of purposeful selection (19), which is based on both theoretical and statistical considerations. Specifically, we decided which potentially confounding variables available in all 3 cohorts to include in our analysis based on our review of the literature and from analyzing data from the OAI as described in our previously published paper (8). Thus, the multivariable knee analysis was adjusted for sex, race, cohort study (i.e., OAI, MOST, or CHECK), and baseline values for the following variables: BMI category; age; walking (seldom, sometimes, or often) or not walking for activity; knee osteoarthritis grade as assessed by radiography (i.e., Kellgren/Lawrence grade [20]); persistent knee pain; number of comorbidities; marriage status; employment status; and education status. The multivariable hip analysis was adjusted for sex, race, cohort study, and baseline values for the following variables: BMI category; age; walking (seldom, sometimes, or often) or not walking for activity; persistent hip pain; and number of comorbidities. More information about the definition of variables can be found in Supplementary Appendix A, available at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021. Note that we added the variable of cohort study (i.e., OAI, MOST, or CHECK) into our analyses as a covariate rather than incorporating it into a random‐effects model, as our meta‐analyses (to be detailed below, and as shown in Supplementary Appendix A) did not show any evidence of heterogeneity between the cohort studies.

In our analyses, we treated our exposure of interest (i.e., change in BMI) as the continuous variable that it is instead of collapsing it into categories (e.g., decrease in BMI, stable BMI, and increase in BMI, as might be the case in a target trial emulation). Categorization of continuous exposure/predictor variables is not recommended in research (21, 22) as it distorts the association between the predictor and outcome (23), causes substantial loss of power and precision (24, 25), reduces the predictive efficiency of the analysis with increased probability of biased estimates (25, 26), may also inflate the rate of Type I errors (false‐positives) (27) or Type II errors (false‐negatives) (28), and can markedly inflate the effect size (odds ratio) (29, 30). However, we used categorized change in BMI (i.e., 3 categories of changes in BMI: decrease in BMI [of ≥1 BMI units], stable BMI [<1 BMI unit decrease or increase], and increase in BMI [of ≥1 BMI units]) when reporting descriptive statistics about the participants/knees or hips in this study (e.g., baseline characteristics).

Our analyses used individual knees and hips as observational units while accounting for the fact that up to 2 knees and 2 hips were clustered in the same participant and were therefore not independent of each other. In order to account for this nonindependence, we used the marginal approach for multivariable failure time data in Cox proportional hazards models (31), in which robust variance–covariance estimators are used to account for the correlation between units within clusters (31).

To assess whether the estimations for any associations between change in BMI and knee or hip replacements were influenced by the BMI category at baseline, age at baseline, and sex, we investigated the possibility of interactions between change in BMI from baseline and each of these 3 variables (i.e., BMI category at baseline, age at baseline, and sex).

We included all available observations in our analyses, and in our primary analyses, no imputation for missing data was performed (see below for details of sensitivity analyses for missing data). We set our threshold for statistical significance as a 2‐tailed P value of less than 0.05.

We tested the assumption of proportionality of hazards in the Cox proportional hazards models by adding the interaction term between each predictor variable × log(time) to the models and checking whether the interaction term had a statistically significant effect (P < 0.05) in the model (32). None were statistically significant. The overall goodness of fit of each model was assessed visually by inspecting the plot of the log cumulative hazard against the Cox‐Snell residuals (33).

We applied 7 types of sensitivity analyses in this study. First, sensitivity analyses were conducted to assess the potential impact of death on the results from the primary analyses. In this type of sensitivity analysis, competing risk analyses were conducted, and the results from the Cox proportional hazards models (our primary analyses) and the competing risk analyses were then compared. In the second type of sensitivity analysis, we checked what the effect might have been if we had added the study cohort (i.e., OAI, MOST, or CHECK) to our analyses as a random effect (i.e., in a multilevel model) rather than as a covariate in the analyses. To this end, we used multilevel mixed‐effects generalized linear models and compared the results with those of our primary analyses. In the third type of sensitivity analysis, in order to assess if using a different statistical method would confirm our findings from the Cox proportional hazards model (our primary analyses), we used generalized estimating equations with a logistic link function (i.e., logistic regression with clustering of up to 2 knees and 2 hips per individuals) (34) to investigate the association between change in BMI (in kg/m2) between baseline and 7–10 years’ follow‐up and knee and hip replacements during that time.

In the fourth type of sensitivity analysis, we additionally removed participants who had hip replacement prior to baseline from the knee cohort, and participants who had knee replacement prior to baseline from the hip cohort, and compared the results with our primary analyses in order to assess any potential impact of altered biomechanics from having a preexisting joint replacement. As a fifth type of sensitivity analysis, we investigated the impact of missing data in the analyses by imputing missing data on change in BMI in 1 of 5 ways (2 SDs less than the mean [worst case scenario]; 1 SD less than the mean [worst best case scenario]; 1 SD more than the mean [best worst case scenario]; and 2 SDs more than the mean [best case scenario]) for data that was missing not at random (MNAR) and multiple imputations for data that were missing at random. Although our statistical analyses showed that data for the knee and hip cohorts were MNAR (and thus were not missing at random), we nonetheless performed multiple imputation for additional scrutiny.

As a sixth type of sensitivity analysis, we meta‐analyzed the results from the 3 study cohorts. This was done for the following 2 purposes: 1) to determine if the overall effect sizes (i.e., hazard ratios [HRs]) obtained from the meta‐analyses were different those obtained in our primary analyses; and 2) to assess heterogeneity between the 3 cohort studies used in the current study. As the seventh (last) type of sensitivity analysis, we assessed what the effect would have been if we had limited our analysis to the outcome of only those knee and hip replacements that were due to osteoarthritis and other degenerative arthritis instead of what we did in our primary analyses, where we analyzed knees and hips that had been replaced for any reason. To conduct this sensitivity analysis, the necessary data (i.e., the reason for knee or hip replacement) were only available in the OAI and MOST study, but not in the CHECK study. Thus, for the CHECK study, we assumed conservatively that all joint replacements were not due to osteoarthritis and other degenerative arthritis.

All statistical analyses, except meta‐analyses, were conducted using Stata software, version 17.0 (basic edition). The “punafcc” package in Stata (35) was used for PAF estimations. The meta‐analyses were conducted using Comprehensive Meta Analysis software, version 3.3.070.

Public and patient involvement

The participants whose data were used for the current study were not directly consulted in the design of this study. The participants were not asked to provide advice on the interpretation of or writing of the results. We are disseminating the main results to the broader community via articles on public‐facing websites. We are open to patient and public involvement in the development of other appropriate methods of dissemination.

RESULTS

Characteristics of the knee and hip cohorts used in this study

Tables 1 and 2 show the baseline characteristics of participants and knees or hips in the knee or hip cohorts, respectively, stratified by change in BMI from baseline: decrease of ≥1 BMI units (i.e., ≥1 kg/m2); <1 unit change in BMI (stable BMI); or increase of ≥1 BMI units. There were 8,181 participants (16,362 knees) and 8,203 participants (16,406 hips) in the knee and hip cohorts, respectively. Of these knees and hips, 548 knees were from 281 participants (of 8,181 participants in total) who had missing data on change in BMI, and 496 hips were from 250 participants (of 8,203 participants in total) who had missing data on change in BMI. Of the knees and hips from participants who did not have missing data for change in BMI, 3,956 (25.0%) and 4,030 (25.3%), respectively, were exposed to a decrease of ≥1 BMI units, while 5,034 (31.8%) and 5,145 (32.3%), respectively, were exposed to an increase of ≥1 BMI units (Tables 1 and 2).

Table 1.

Baseline characteristics of participants and knees in the knee cohort stratified by change in body mass index (BMI) from baseline*

Characteristic Decrease in BMI (of ≥1 BMI units) Stable BMI (<1 BMI‐unit decrease or increase) Increase in BMI (of ≥1 BMI units) Total P
Participants 1,982 (25.1) 3,397 (43.0) 2,521 (31.9) 7,900 (100.0) –
Age, mean ± SD years 61.6 ± 8.6 61.3 ± 8.7 58.5 ± 7.8 60.5 ± 8.5 <0.01
Sex <0.01
Female 1,262 (63.7) 1,982 (58.4) 1,636 (64.9) 4,880 (61.8) –
Male 720 (36.3) 1,415 (41.7) 885 (35.1) 3,020 (38.2) –
Race 0.38
White 1,616 (81.5) 2,796 (82.3) 2,097 (83.2) 6,509 (82.4) –
Other 366 (18.5) 601 (17.7) 424 (16.8) 1,391 (17.6) –
BMI, mean ± SD kg/m2 30.5 ± 5.6 28.5 ± 5.0 28.9 ± 5.2 29.1 ± 5.3 <0.01
BMI category <0.01
Underweight (<18.5 kg/m2) (excluded) 0 (0) 0 (0) 0 (0) 0 (0)
Normal (18.5 kg/m2 ≤BMI <25.0 kg/m2) 294 (14.8) 873 (25.7) 589 (23.4) 1,756 (22.2) –
Overweight (25.0 kg/m2 ≤BMI <30.0 kg/m2) 718 (36.2) 1,380 (40.6) 971 (38.5) 3,069 (38.9) –
Obese (30.0 kg/m2 ≤BMI) 970 (49.0) 1,144 (33.7) 961 (38.1) 3,075 (38.9) –
Walking or not walking for physical activity 0.36
Never 238 (12.0) 452 (13.3) 314 (12.5) 1,004 (12.7) –
Seldom, sometimes or often 1,740 (88.0) 2,942 (86.7) 2,199 (87.5) 6,881 (87.3) –
Marital status <0.01
Never married 145 (7.4) 199 (5.9) 191 (7.6) 535 (6.8) –
Was with a partner (widowed, divorced, or separated) 468 (23.8) 711 (21.1) 554 (22.2) 1,733 (22.1) –
Married 1,355 (68.8) 2,464 (73.0) 1,757 (70.2) 5,576 (71.1) –
No. of comorbidities <0.01
0 1,266 (64.2) 2,444 (72.5) 1,749 (70.1) 5,459 (69.6) –
1 401 (20.3) 598 (17.7) 444 (17.8) 1,443 (18.4) –
≥2 306 (15.5) 331 (9.8) 303 (12.1) 940 (12.0) –
Employment status at baseline <0.01
Working 1,099 (55.5) 1,976 (58.2) 1,625 (64.6) 4,700 (59.6) –
Not working 880 (44.5) 1,416 (41.8) 890 (35.4) 3,186 (40.4) –
Education status at baseline 0.01
Above high school 1,410 (71.5) 2,543 (75.2) 1,854 (74.0) 5,807 (73.9) –
High school or below 561 (28.5) 839 (24.8) 651 (26.0) 2,051 (26.1) –
Cohort study <0.01
Osteoarthritis Initiative study 1,000 (50.4) 1,906 (56.1) 1,281 (50.8) 4,187 (53.0) –
Multicenter Osteoarthritis study 285 (14.4) 239 (7.1) 381 (15.1) 905 (11.5) –
Cohort Hip and Cohort Knee study 697 (35.2) 1,252 (36.8) 859 (34.1) 2,808 (35.5) –
Knees 3,956 (25.0) 6,824 (43.2) 5,034 (31.8) 15,814 (100.0) –
Kellgren/Lawrence grade <0.01
None (grade 0) 1,184 (30.2) 2,071 (30.5) 1,844 (36.9) 5,099 (32.5) –
Doubtful OA (grade 1) 1,122 (28.6) 1,914 (28.2) 1,407 (28.1) 4,443 (28.3) –
Minimal OA (grade 2) 732 (18.6) 1,139 (16.8) 835 (16.7) 2,706 (17.2) –
Moderate OA (grade 3) 683 (17.4) 1,219 (17.9) 715 (14.3) 2,617 (16.6) –
Severe OA (grade 4) 205 (5.2) 445 (6.6) 198 (4.0) 848 (5.4) –
Persistent knee pain at baseline 0.26
No 1,940 (54.1) 3,395 (55.5) 2,451 (54.1) 7,786 (54.7) –
Yes 1,643 (45.9) 2,720 (44.5) 2,077 (45.9) 6,440 (45.3) –
Open in a new tab
*

Values are the number (%) unless indicated otherwise. The percentage calculations are based on complete case (i.e., excluding missing values). Chi‐square and Kruskal‐Wallis test analyses were used for comparisons between BMI change groups. OA = osteoarthritis.

Table 2.

Baseline characteristics of participants and hips in the hip cohort stratified by change in body mass index (BMI) from baseline*

Characteristic Decrease in BMI (of ≥1 BMI units) Stable BMI (<1 BMI‐unit decrease or increase) Increase in BMI (of ≥1 BMI units) Total P
Participants 2,017 (25.4) 3,362 (42.3) 2,574 (32.3) 7,953 (100.0) –
Age, mean ± SD years 61.7 ± 8.6 61.3 ± 8.7 58.6 ± 7.8 60.5 ± 8.5 <0.01
Sex <0.01
Female 1,293 (64.1) 1,956 (58.2) 1,681 (65.3) 4,930 (62.0) –
Male 724 (35.9) 1,406 (41.8) 893 (34.7) 3,023 (38.0) –
Race 0.39
White 1,642 (81.4) 2,760 (82.1) 2,135 (82.9) 6,537 (82.2) –
Other 375 (18.6) 602 (17.9) 439 (17.1) 1,416 (17.8) –
BMI, mean ± SD kg/m2 30.6 ± 5.7 28.4 ± 5.0 29.0 ± 5.3 29.2 ± 5.4 <0.01
BMI category <0.01
Underweight (<18.5 kg/m2) (excluded) 0 (0) 0 (0) 0 (0) 0 (0)
Normal (18.5 kg/m2 ≤BMI <25.0 kg/m2) 293 (14.5) 863 (25.7) 592 (23.0) 1,748 (22.0) –
Overweight (25.0 kg/m2 ≤BMI <30.0 kg/m2) 712 (35.3) 1,378 (41.0) 992 (38.5) 3,082 (38.7) –
Obese (30.0 kg/m2 ≤BMI) 1,012 (50.2) 1,121 (33.3) 990 (38.5) 3,123 (39.3) –
Walking or not walking for physical activity 0.69
Never 248 (12.3) 438 (13.0) 320 (12.5) 1,006 (12.7) –
Seldom, sometimes or often 1,765 (87.7) 2,920 (87.0) 2,247 (87.5) 6,932 (87.3) –
No. of comorbidities <0.01
0 1,281 (63.8) 2,423 (72.6) 1,788 (70.1) 5,492 (69.6) –
1 411 (20.5) 595 (17.8) 446 (17.5) 1,452 (18.4) –
≥2 316 (15.7) 320 (9.6) 315 (12.4) 951 (12.0) –
Cohort study <0.01
Osteoarthritis Initiative study 1,010 (50.1) 1,896 (56.4) 1,294 (50.3) 4,200 (52.8) –
Multicenter Osteoarthritis study 278 (13.8) 246 (7.3) 377 (14.6) 901 (11.3) –
Cohort Hip and Cohort Knee study 729 (36.1) 1,220 (36.3) 903 (35.1) 2,852 (35.9) –
Hips 4,030 (25.3) 6,735 (42.4) 5,145 (32.3) 15,910 (100.0) –
Persistent hip pain at baseline <0.01
No 2,234 (72.0) 3,903 (75.4) 2,923 (73.6) 9,060 (73.9) –
Yes 869 (28.0) 1,274 (24.6) 1,048 (26.4) 3,191 (26.1) –
Open in a new tab
*

Values are the number (%) unless indicated otherwise. The percentage calculations are based on complete case (i.e., excluding missing values). Chi‐square and Kruskal‐Wallis test analyses were used for comparisons between BMI change groups.

There were some significantly different characteristics between participants in these 3 stratified groups of change in BMI in both the knee and hip cohorts, notably for age at baseline, BMI at baseline, and sex. Specifically, participants who had a decrease of ≥1 BMI units were the oldest on average (mean ± SD 61.6 ± 8.6 and 61.7 ± 8.6 years in the knee and hip cohorts, respectively) and had the highest BMI at baseline (mean ± SD 30.5 ± 5.6 and 30.6 ± 5.7 kg/m2 in the knee and hip cohorts, respectively). The group of participants who had an increase of ≥1 BMI units had the highest percentage of female participants (64.9% and 65.3% in the knee and hip cohorts, respectively). The median follow‐up time was 7.0 years (interquartile range 6.0–8.0 years) in both the knee and hip cohorts. Detailed information on baseline characteristics of the knee and hip cohorts can be found in Supplementary Tables 1–2, available on the Arthritis Care & Research website at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021.

Figure 2 shows the change in BMI from baseline to the last follow‐up visit before knee or hip replacement for participants in the knee and hip cohorts, or for participants who did not have a knee or hip replacement, the last follow‐up visit. For participants who had 2 knee or 2 hip replacements during the study, the first knee or first hip replacement was used in the calculation of change in BMI in Figure 2. There were nonnegligible minorities of participants who had a change in BMI of ≥5 units (which is the maximum amount required to change from some BMI categories, such as from the overweight to the normal category). Specifically, in the knee cohort, 2.1% of participants had a decrease and 5.7% of participants had an increase of ≥5 BMI units, while in the hip cohort, 2.4% of participants had a decrease and 5.5% of participants had an increase of ≥5 BMI units.

Figure 2.

Figure 2

Open in a new tab

Histogram of change in body mass index (BMI; kg/m2) from baseline in participants in the knee (A) and hip (B) cohorts.

Knee replacement

Of the 16,362 knees from 8,181 participants in the knee cohort, there were 999 knee replacements (6.1%). Of those 999 knee replacements, 198 (19.8%) occurred in participants who had a decrease of ≥1 BMI units; 260 (26.0%) occurred in participants who had an increase of ≥1 BMI units; 415 (41.6%) occurred in participants who had a <1–unit change in BMI (stable BMI); and 126 (12.6%) occurred in participants who had missing BMI data at the relevant follow‐up before knee replacement, so their change in BMI from baseline could not be calculated.

Table 3 shows the association between change in BMI from baseline (in kg/m2) and the risk of knee replacement as determined by the Cox proportional hazards models in unadjusted (univariate) and adjusted (multivariable) models. While the unadjusted Cox proportional hazards model showed no significant association between change in BMI and knee replacement, the adjusted model showed a significant association between change in BMI from baseline and the risk of knee replacement. Indeed, the adjusted HR (HRadj) was 1.03 (95% confidence interval [95% CI] 1.00–1.06), suggesting that each 1‐unit decrease in BMI from baseline was associated with a 3% decreased risk of knee replacement. There were no significant interactions between change in BMI from baseline and BMI category at baseline, age at baseline, or sex. Supplementary Table 3, available on the Arthritis Care & Research website at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021, shows the associations between change in BMI and the risk of knee replacement for each study cohort.

Table 3.

Association between change in body mass index (BMI) from baseline until knee or hip replacement (or until 7–10 years’ follow‐up for knees or hips that were not replaced) and the risk of knee or hip replacement, as shown in univariate and multivariable analysis*

Outcome Univariate analysis Multivariable analysis†
HR (95% CI) P HR (95% CI) P
Knee replacement 1.00 (0.97–1.03) 0.81 1.03 (1.00–1.06) 0.022‡
Hip replacement 0.98 (0.94–1.01) 0.22 1.00 (0.95–1.04) 0.92
Open in a new tab
*

95% CI = 95% confidence interval; HR = hazard ratio.

†

Multivariable knee analysis was adjusted for sex, race, cohort study (i.e., Osteoarthritis Initiative, Multicenter Osteoarthritis study, or Cohort Hip and Cohort Knee study), and baseline values for the following variables: BMI category; age; walking (seldom, sometimes, or often) or not walking for activity; knee osteoarthritis grade as assessed by radiography (i.e., Kellgren/Lawrence grade); persistent knee pain; number of comorbidities; marriage status; employment status; and education status. The multivariable hip analysis was adjusted for sex, race, cohort study, and baseline values for the following variables: BMI category; age; walking (seldom, sometimes, or often) or not walking for activity; persistent hip pain; and number of comorbidities.

‡

Significant.

Hip replacement

Of the 16,406 hips from 8,203 participants in the hip cohort, there were 341 hip replacements (2.1%). Of those 341 hip replacements, 77 (22.6%) occurred in participants who had a decrease of ≥1 BMI units; 82 (24.0%) occurred in participants who had an increase of ≥1 BMI units; 125 (36.7%) occurred in participants who had a <1–unit change in BMI (stable BMI); and 57 (16.7%) occurred in participants who had missing data for calculating change in BMI from baseline.

Table 3 shows the association between change in BMI from baseline (in kg/m2) and the risk of hip replacement as determined by the Cox proportional hazards models in unadjusted (univariate) and adjusted (multivariable) models. The unadjusted and adjusted Cox proportional hazards models for hips showed no significant association between change in BMI and hip replacement. The HRadj was 1.00 (95% CI 0.95–1.04) (Table 3). There were no significant interactions between change in BMI from baseline and any of the following variables: BMI category at baseline, age at baseline, or sex. Supplementary Table 3, available at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021, shows the associations between change in BMI and the risk of hip replacement for each study cohort.

Survival plots

Figure 3 shows the survival plots for knees and hips, with the probability of a knee or hip surviving (i.e., avoiding replacement), using the averages for all the variables included in the Cox proportional hazards models shown in Table 3 for 3 different changes in BMI: a 5‐unit decrease in BMI; a 5‐unit increase in BMI; and stable BMI (i.e., 0‐unit change in BMI). We used a 5‐unit decrease or increase in BMI for the survival plots for ease of visualization and because 5 units of change in BMI is the minimum required to change BMI category (e.g., from overweight to normal, or from overweight to obese). The knees that were exposed to a 5‐unit decrease in BMI consistently had a higher probability of avoiding replacement than those exposed to stable BMI or a 5‐unit increase in BMI. For hips, there was no effect of any change in BMI on the probability of avoiding hip replacement, which was as expected, given the results for the hip mentioned above.

Figure 3.

Figure 3

Open in a new tab

Survival plots for knees (A) and hips (B) (i.e., the probability of avoiding replacement over follow‐up time). The probability of avoiding knee and hip replacement was calculated using the averages for the variables included in the Cox proportional hazards multivariable models shown in Table 3 and all the coefficients from these models in the 3 categories of change in body mass index (BMI) (decrease in BMI, stable BMI, and increase in BMI) shown on the figure.

PAFs

The calculation of PAFs using the adjusted Cox proportional hazards models showed evidence that a 1‐unit decrease in BMI (assuming nothing else changed) could reduce the incidence of knee replacement by 3.2% (range 0.3–5.9%) in individuals with a BMI ≥25 kg/m2 (i.e., those with overweight or obesity) and by 4.9% (range 0.3–9.2%) in individuals with a BMI of 18.5 to <25 kg/m2 (i.e., those with a BMI in the normal range). If we apply this calculation to all individuals with a BMI ≥18.5 kg/m2 (i.e., a BMI in the normal, overweight, or obese categories combined), then a 1‐unit decrease in BMI could reduce the incidence of knee replacement by 3.4% (range 0.7–6.0%). In hips, there was no significant reduction in the incidence of hip replacements in any BMI categories, as expected based on the results from the Cox proportional hazards model (Table 3).

Sensitivity analyses

In our 7 types of sensitivity analyses, the results we obtained were either identical or similar to the results we obtained in our primary analyses for the knee and hip using Cox proportional hazards models (Table 3), thus validating the results from our primary analyses. Detailed results from our 7 types of sensitivity analyses can be found in Supplementary Appendix A, available at http://onlinelibrary.wiley.com/doi/10.1002/acr.25021. Noteworthy is the fact that while our sensitivity analyses investigating the impact of data MNAR showed significant associations between change in BMI and hip replacement when missing data were imputed in the 4 different ways (i.e., worst case scenario; worst best case scenario; best worst case scenario; and best case scenario), the associations were in opposite directions in the worst and worst best case scenarios compared to the best worst and best case scenarios. These inconsistent findings for the hip provide further evidence that there is no clear association of change in BMI with hip replacement.

DISCUSSION

This large, multicohort, long‐term longitudinal study (8,181–8,203 participants from 3 cohorts with 7–10 years’ follow‐up) found that, among individuals with or at risk of clinically significant knee osteoarthritis, there was an association between change in BMI from baseline and the risk of knee replacement, regardless of baseline BMI. In contrast, no evidence was found for an association between change in BMI and hip replacement. This suggests that weight loss may reduce the risk of, or delay, the need for knee but not hip replacement, while weight gain may increase the risk of, or accelerate, the need for knee but not hip replacement.

It is noteworthy that the benefit of weight loss for reducing the need for knee replacement in this multicohort study was not only seen in individuals with overweight or obesity (i.e., a BMI ≥25 kg/m2), but also in those of normal BMI (i.e., a BMI of 18.5 to <25 kg/m2), as was also shown in our previous, single‐cohort study (8). This association between decrease in BMI and knee replacement in individuals in a range of BMI categories could potentially have a sizeable impact on public health and provide important cost shifts. Indeed, our estimates showed that if all knees from the population of individuals with or at risk of clinically significant knee osteoarthritis and with a BMI ≥18.5 kg/m2 (i.e., a BMI in the normal, overweight, or obese range) were exposed to a decrease in BMI of 1 unit (i.e., 1 kg/m2), the risk of knee replacement could be reduced by 3.4% (range 0.7–6.0%). Weight loss is recommended for patients with osteoarthritis and overweight or obesity in current international clinical guidelines (36, 37, 38, 39, 40, 41). If this recommendation for weight loss were extended in the guidelines to individuals of normal BMI with or at risk of clinically significant knee osteoarthritis but who are otherwise healthy, these patients could potentially benefit from a reduced need for knee replacement. However, the potential benefits of weight loss for knee osteoarthritis in individuals of normal BMI would need to be weighed against the potential dangers of weight loss in this population, particularly for older individuals (8).

With the current study, there are now 4 studies that have investigated the association between change in BMI and knee and hip replacement (8, 9, 10). Of these 4 studies, 3 studies (the current study and those of Salis et al and Jin et al [8, 9]) found an association between weight loss and reduced risk of knee replacement, whereas the last of the 4 studies did not find any such association (10). It is noteworthy that the first 3 of these 4 studies all had a follow‐up time of 5.2–10.0 years, whereas the last of the 4 studies (10) had a follow‐up time of 4 years. It may be that 4 years is of insufficient duration to show any association between weight loss and knee replacement. In contrast to the knee, only 1 of these 4 studies (8) (from the OAI cohort with a 10‐year follow‐up) found an association between weight loss and reduced risk of hip replacement, while the remaining 3 of these 4 studies (the current study from the OAI, MOST, and CHECK cohorts and the studies of Jin et al and Joseph et al [9, 10] from the OAI cohort and an Australian population‐based cohort, with follow‐up times of 4–10 years) did not find any such association. The divergent finding for hip replacement in the study mentioned above (8) is likely due to differences in overall cohort composition and follow‐up time. When combining all results from the above 4 studies, it appears that weight loss has a robust association with a reduced risk of knee replacement across different cohorts but is not robustly associated with a reduced risk of hip replacement.

The difference between the knee and hip with respect to the impact of change in weight (or BMI) on knee and hip replacement is in keeping with a previous suggestion that the knee joint may be more sensitive to weight than the hip joint (42). Possible reasons for relative resistance of the hip to effects of weight are 1) the ball‐and‐socket form of the hip reduces the vertical loading in gait by distributing it across the joint, in comparison to the hinge form of the knee (43, 44); 2) the hip is less susceptible to metabolic damage of bone and cartilage in comparison to the knee (42); and 3) the cartilage of the hip is markedly thicker than the cartilage of the knee (45).

This study has several limitations. First, it has the inherent bias and confounding associated with observational studies. Future randomized controlled trials (RCTs) are required to determine any potential causality of weight loss in reducing the incidence of knee replacement. Setting up such long‐term RCTs for knee replacement is challenging because osteoarthritis develops and progresses slowly (46). However, surrogate outcomes instead of knee replacement determining end‐stage knee osteoarthritis (47) could be used in such RCTs in the future. Second, there were likely latent confounders that were not captured in the 3 cohort studies used in our analyses. Third, participants in this study were primarily those with or at risk of clinically significant knee osteoarthritis; therefore, the current findings on hip replacement may not be transferable to individuals with hip osteoarthritis unless they also have or are at risk of clinically significant knee osteoarthritis. Last, our study cohorts were mainly White participants and elderly. Therefore, the transferability of our findings to populations with different characteristics is limited.

In conclusion, weight loss holds the potential to reduce the risk of, or delay the need for, knee but not hip replacement for individuals with overweight or obesity and for those of normal weight. Moreover, weight loss thus holds the potential to reduce the health burden of knee but not hip replacement surgery.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Mr. Salis had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design

Salis, Sainsbury.

Acquisition of data

Salis.

Analysis and interpretation of data

Salis, Sainsbury.

Supporting information

Disclosure Form

ACR-75-1340-s003.pdf (1.3MB, pdf)

Supplementary Appendix A

ACR-75-1340-s002.docx (69.6KB, docx)

Supplementary Table 1. Baseline characteristics of participants and knees in the knee cohort from each study (OAI, MOST or CHECK), stratified by change from baseline in body mass index

Supplementary Table 2. Baseline characteristics of participants and hips in the hip cohort from each study (OAI, MOST or CHECK), stratified by change from baseline in body mass index

Supplementary Table 3. Association between change in body mass index from baseline until knee or hip replacement – or until 7 to 10 years’ follow up for knees or hips that were not replaced – and the risk of knee and hip replacement, as shown in univariate and multivariable analysis for each study cohort

ACR-75-1340-s001.docx (94.3KB, docx)

ACKNOWLEDGMENTS

We acknowledge the contributions of the study participants, investigators, research staff involved, and the provision of data sets and/or research tools from 3 cohort studies: the OAI, the MOST study, and the CHECK study. Open access publishing facilitated by University of New South Wales, as part of the Wiley ‐ University of New South Wales agreement via the Council of Australian University Librarians.

The content herein was prepared using Multicenter Osteoarthritis Study (MOST) data and does not claim, infer, or imply endorsement by MOST, by the MOST investigators and their respective institutions, or by the University of California of the data recipients’ use of the data, of the entity or personnel conducting the research, or of any results of the current study.

This article was prepared using an Osteoarthritis Initiative (OAI) public‐use data set, and its contents do not necessarily reflect the opinions or views of the OAI Study Investigators, the NIH, or the private funding partners of the OAI. The OAI is a public–private partnership between the NIH (contracts N01‐AR‐2‐2258, N01‐AR‐2‐2259, N01‐AR‐2‐2260, N01‐AR‐2‐2261, and N01‐AR‐2‐2262) and private funding partners (Merck Research Laboratories, Novartis Pharmaceuticals, GlaxoSmithKline, and Pfizer, Inc.) and is conducted by the OAI Study Investigators. Private sector funding for the OAI is managed by the Foundation for the NIH. The authors of this article are not part of the OAI investigative team.

The MOST study is supported by 4 cooperative grants: U01‐AG‐18820 to David T. Felson at Boston University; U01‐AG‐18832 to James Torner at University of Iowa; U01‐AG‐18947 to Cora E. Lewis at the University of Alabama at Birmingham; and U01‐AG‐19069 to Michael C. Nevitt at the University of California, San Francisco. It is funded by the NIH, a branch of the Department of Health and Human Services, and conducted by MOST investigators. The Cohort Hip and Cohort Knee (CHECK) study is funded by the Dutch Arthritis Foundation. The following centers are involved: Erasmus Medical Center Rotterdam; Kennemer Gasthuis Haarlem; Leiden University Medical Center; Maastricht University Medical Center; Martini Hospital Groningen/Allied Health Care Center for Rheumatology and Rehabilitation Groningen; Medical Spectrum Twente Enschede/Ziekenhuisgroep Twente Almelo; Reade Center for Rehabilitation and Rheumatology; St Maartens‐kliniek Nijmegen; University Medical Center Utrecht; and Wilhelmina Hospital Assen. Mr. Salis is recipient of an Australian Government Research Training Program Scholarship. Dr. Sainsbury's work was supported by the National Health and Medical Research Council of Australia (Senior Research fellowship 1135897).

REFERENCES

  • 1. Hunter DJ, Schofield D, Callander E. The individual and socioeconomic impact of osteoarthritis. Nat Rev Rheumatol 2014;10:437–41. [DOI] [PubMed] [Google Scholar]
  • 2. Leifer VP, Katz JN, Losina E. The burden of OA‐health services and economics. Osteoarthritis Cartilage 2022;30:10–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Australian Institute of Health and Welfare . Health‐care expenditure on arthritis and other musculoskeletal conditions 2008–09. Canberra: AIHW, Australian Government; 2014. [Google Scholar]
  • 4. Hunter DJ, Bierma‐Zeinstra S. Osteoarthritis. Lancet 2019;393:1745–59. [DOI] [PubMed] [Google Scholar]
  • 5. Bliddal H, Leeds AR, Christensen R. Osteoarthritis, obesity and weight loss: evidence, hypotheses and horizons: a scoping review. Obes Rev 2014;15:578–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Muthuri SG, Hui M, Doherty M, et al. What if we prevent obesity? Risk reduction in knee osteoarthritis estimated through a meta‐analysis of observational studies. Arthritis Care Res (Hoboken) 2011;63:982–90. [DOI] [PubMed] [Google Scholar]
  • 7. Lohmander LS, Gerhardsson de Verdier M, Rollof J, et al. Incidence of severe knee and hip osteoarthritis in relation to different measures of body mass: a population‐based prospective cohort study. Ann Rheum Dis 2009;68:490–6. [DOI] [PubMed] [Google Scholar]
  • 8. Salis Z, Sainsbury A, Keen HI, et al. Weight loss is associated with reduced risk of knee and hip replacement: a survival analysis using Osteoarthritis Initiative data. Int J Obes (Lond) 2022;46:874–84. [DOI] [PubMed] [Google Scholar]
  • 9. Jin X, Gibson AA, Gale J, et al. Does weight loss reduce the incidence of total knee and hip replacement for osteoarthritis? A prospective cohort study among older adults with overweight or obesity. Osteoarthritis Cartilage 2020;28 Supplement 1:S419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Joseph GB, McCulloch CE, Nevitt MC, et al. Effects of weight change on knee and hip radiographic measurements and pain over four years: data from the Osteoarthritis Initiative. Arthritis Care Res (Hoboken) 2023;75:860–68. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. National Institutes of Health . Osteoarthritis Initiative (OAI) study protocol. URL: https://nda.nih.gov/oai/study_documentation.html.
  • 12. Segal NA, Nevitt M, Gross KD, et al. The multicenter osteoarthritis study: opportunities for rehabilitation research. PM R 2013;5:647–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Wesseling J, Boers M, Viergever MA, et al. Cohort profile: Cohort Hip and Cohort Knee (CHECK) study. Int J Epidemiol 2016;45:36–44. [DOI] [PubMed] [Google Scholar]
  • 14. Lui M, Jones CA, Westby MD. Effect of non‐surgical, non‐pharmacological weight loss interventions in patients who are obese prior to hip and knee arthroplasty surgery: a rapid review. Syst Rev 2015;4:121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Godziuk K, Prado CM, Beaupre L, et al. A critical review of weight loss recommendations before total knee arthroplasty. Joint Bone Spine 2021;88:105114. [DOI] [PubMed] [Google Scholar]
  • 16. Pellegrini CA, Ledford G, Hoffman SA, et al. Preferences and motivation for weight loss among knee replacement patients: implications for a patient‐centered weight loss intervention. BMC Musculoskeletal Disord 2017;18:327. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Coggon D, Rose G, Barker DJ. Epidemiology for the uninitiated. Measurement error and bias. URL: https://www.bmj.com/about‐bmj/resources‐readers/publications/epidemiology‐uninitiated/4‐measurement‐error‐and‐bias. [Google Scholar]
  • 18. Mansournia MA, Altman DG. Population attributable fraction. BMJ 2018;360:k757. [DOI] [PubMed] [Google Scholar]
  • 19. Lemeshow S, May S, Hosmer DW Jr. Applied survival analysis: regression modeling of time‐to‐event data. 2nd ed. 2011. Hoboken (NJ): John Wiley & Sons; 2011. [Google Scholar]
  • 20. Kellgren JH, Lawrence JS. Radiological assessment of osteo‐arthrosis. Ann Rheum Dis 1957;16:494–502. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med 2006;25:127–41. [DOI] [PubMed] [Google Scholar]
  • 22. Altman DG. Categorizing continuous variables. In: Wiley StatsRef: Statistics Reference Online. 2014. URL: 10.1002/9781118445112.stat04857. [DOI] [Google Scholar]
  • 23. Nguyen TV. Common methodological issues and suggested solutions in bone research. Osteoporos Sarcopenia 2020;6:161–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Faraggi D, Simon R. A simulation study of cross‐validation for selecting an optimal cutpoint in univariate survival analysis. Stat Med 1996;15:2203–13. [DOI] [PubMed] [Google Scholar]
  • 25. Taylor JM, Yu M. Bias and efficiency loss due to categorizing an explanatory variable. J Multivar Anal 2002;83:248–63. [Google Scholar]
  • 26. Becher H. The concept of residual confounding in regression models and some applications. Stat Med 1992;11:1747–58. [DOI] [PubMed] [Google Scholar]
  • 27. Austin PC, Brunner LJ. Inflation of the type I error rate when a continuous confounding variable is categorized in logistic regression analyses. Stat Med 2004;23:1159–78. [DOI] [PubMed] [Google Scholar]
  • 28. Gupta R, Day CN, Tobin WO, et al. Understanding the effect of categorization of a continuous predictor with application to neuro‐oncology. NeuroOncol Pract 2021;9:87–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Naggara O, Raymond J, Guilbert F, et al. Analysis by categorizing or dichotomizing continuous variables is inadvisable: an example from the natural history of unruptured aneurysms. AJNR Am J Neuroradiol 2011;32:437–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. O'Gorman TW, Woolson RF. The effect of category choice on the odds ratio and several measures of association in case‐control studies. Commun Stat Theory Methods 1993;22:1157–71. [Google Scholar]
  • 31. Lin DY. Cox regression analysis of multivariate failure time data: the marginal approach. Stat Med 1994;13:2233–47. [DOI] [PubMed] [Google Scholar]
  • 32. Cleves M, Gould WW, Marchenko YV. An introduction to survival analysis using STATA, revised. 3rd ed. College Station (TX): Stata Press; 2016. [Google Scholar]
  • 33. Cox DR, Snell EJ. A general definition of residuals. J Royal Stat Soc Series B (Methodological) 1968;30:248–75. [Google Scholar]
  • 34. Ballinger GA. Using generalized estimating equations for longitudinal data analysis. Organizational Res Methods 2004;7:127–50. [Google Scholar]
  • 35. Newson RB. Attributable and unattributable risks and fractions and other scenario comparisons. Stata J 2013;13:672–98. [Google Scholar]
  • 36. Royal Australian College of General Practitioners (RACGP) . Guideline for the management of knee and hip osteoarthritis. 2nd ed. 2018. URL: https://www.racgp.org.au/download/Documents/Guidelines/Musculoskeletal/guideline‐for‐the‐management‐of‐knee‐and‐hip‐oa‐2nd‐edition.pdf.
  • 37. Kolasinski SL, Neogi T, Hochberg MC, et al. 2019 American College of Rheumatology/Arthritis Foundation guideline for the management of osteoarthritis of the hand, hip, and knee. Arthritis Care Res (Hoboken) 2020;72:149–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Fernandes L, Hagen KB, Bijlsma JW, et al. EULAR recommendations for the non‐pharmacological core management of hip and knee osteoarthritis. Ann Rheum Dis 2013;72:1125–35. [DOI] [PubMed] [Google Scholar]
  • 39. Jevsevar DS, Brown Ga, Jones DL, et al. The American Academy of Orthopaedic Surgeons evidence‐based guideline on: treatment of osteoarthritis of the knee, 2nd edition. J Bone Joint Surg Am 2013;95:1885–6. [DOI] [PubMed] [Google Scholar]
  • 40. McAlindon TE, Bannuru RR, Sullivan MC, et al. OARSI guidelines for the non‐surgical management of knee osteoarthritis. Osteoarthritis Cartilage 2014;22:363–88. [DOI] [PubMed] [Google Scholar]
  • 41. Brosseau L, Wells GA, Tugwell P, et al. Ottawa Panel evidence‐based clinical practice guidelines for the management of osteoarthritis in adults who are obese or overweight. Phys Ther 2011;91:843–61. [DOI] [PubMed] [Google Scholar]
  • 42. Singer SP, Dammerer D, Krismer M, et al. Maximum lifetime body mass index is the appropriate predictor of knee and hip osteoarthritis. Arch Orthop Trauma Surg 2018;138:99–103. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Reijman M, Pols HA, Bergink AP, et al. Body mass index associated with onset and progression of osteoarthritis of the knee but not of the hip: the Rotterdam Study. Ann Rheum Dis 2007;66:158–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Pamukoff DN, Dudley RI, Vakula MN, et al. An evaluation of the heel strike transient in obese young adults during walking gait. Gait Posture 2016;49:181–3. [DOI] [PubMed] [Google Scholar]
  • 45. Shepherd DE, Seedhom BB. Thickness of human articular cartilage in joints of the lower limb. Ann Rheum Dis 1999;58:27–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Runhaar J, Bierma‐Zeinstra SM. The challenges in the primary prevention of osteoarthritis. Clin Geriatr Med 2022;38:259–71. [DOI] [PubMed] [Google Scholar]
  • 47. Driban JB, Price LL, Lynch J, et al. Defining and evaluating a novel outcome measure representing end‐stage knee osteoarthritis: data from the Osteoarthritis Initiative. Clin Rheumatol 2016;35:2523–30. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Disclosure Form

ACR-75-1340-s003.pdf (1.3MB, pdf)

Supplementary Appendix A

ACR-75-1340-s002.docx (69.6KB, docx)

Supplementary Table 1. Baseline characteristics of participants and knees in the knee cohort from each study (OAI, MOST or CHECK), stratified by change from baseline in body mass index

Supplementary Table 2. Baseline characteristics of participants and hips in the hip cohort from each study (OAI, MOST or CHECK), stratified by change from baseline in body mass index

Supplementary Table 3. Association between change in body mass index from baseline until knee or hip replacement – or until 7 to 10 years’ follow up for knees or hips that were not replaced – and the risk of knee and hip replacement, as shown in univariate and multivariable analysis for each study cohort

ACR-75-1340-s001.docx (94.3KB, docx)

Articles from Arthritis Care & Research are provided here courtesy of Wiley

RESOURCES