To determine if a dose-response relationship exists between percentage changes in body weight in persons with symptomatic knee osteoarthritis (OA) and self-reported pain and function.
To determine if a dose-response relationship exists between percentage changes in body weight in persons with symptomatic knee osteoarthritis (OA) and self-reported pain and function.
Data from persons in the Osteoarthritis Initiative (OAI) and the Multicenter Osteoarthritis (MOST) study data sets (n = 1,410) with symptomatic function-limiting knee OA were studied. For the OAI, we used baseline and 3-year followup data, while for the MOST study, baseline and 30-month data were used. Key outcome variables were Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) physical function and pain change scores. In addition to covariates, the predictor variable of interest was the extent of weight change over the study period divided into 5 categories representing different percentages of body weight change.
A significant dose-response relationship (P < 0.003) was found between the extent of percentage change in body weight and the extent of change in WOMAC physical function and WOMAC pain scores. For example, persons who gained ≥10% of body weight had WOMAC physical function score changes of −5.4 (95% confidence interval −8.7, −2.00) points, indicating worsening physical function relative to the reference group of persons with weight changes between <5% weight gain and <5% weight reduction.
Our data suggest a dose-response relationship exists between changes in body weight and corresponding changes in pain and function. The threshold for this response gradient appears to be body weight shifts of ≥10%. Weight changes of ≥10% have the potential to lead to important changes in pain and function for patient groups as well as individual patients.
Osteoarthritis (OA) of the knee has multiple causes, but one of the more powerful risk factors for OA onset and progression is excessive body weight (1, 2). The Framingham study, for example, reported that women who lost at least 5 kg had a 50% reduction in the odds of developing symptomatic knee OA (3). Given the high costs and high prevalence of knee OA, many researchers have focused on attempts to identify interventions that reduce the body weight of persons with OA who are overweight or obese (4–10).
A meta-analysis that examined the effects of various approaches to weight reduction with or without cointerventions for persons with symptomatic knee OA found that a weight reduction of 5% of body weight was associated with insignificant reductions in knee pain, but significant although small improvements in self-reported functional status (6). Christensen and colleagues also reported a dose-response effect such that the extent of weight reduction was proportional to the extent of functional improvement. In recently published trials, weight reduction strategies leading to losses approximating 10% or more of body weight have resulted in more substantial reductions in pain and improved function (4, 7, 10). In a recently published cohort study of 44 persons undergoing gastric surgery for severe obesity, the mean reduction in body weight from baseline to 6 months following surgery was 20.2% (11). Mean Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain and physical function scores were reduced by 50% or more. These data, in combination, suggest that weight reduction and improvements in pain and functional status may be proportional and respond in a dose-response manner.
Several trials have examined the influences of weight loss on pain and function, and we found one cohort study that examined the effects of body weight gain on pain or functional status (12). If a dose-response relationship exists between body weight changes and corresponding changes in pain and function for persons with symptomatic knee OA, one would expect that body weight gains may also be associated with proportional increases in pain and worsening functional status.
We found no studies that determined if a dose-response relationship existed between body weight changes (both gains and losses) and changes in knee-related pain and functional status in a large sample of persons with symptomatic knee OA. Trial evidence suggests that weight reduction of at least 5% of body weight would lead to improved function and that weight reductions of 10% or more would lead to greater reductions in pain and substantially improved function. The recommendations based on trial findings of persons with knee OA are similar to federal government–based recommendations for weight reduction to optimize health (13, 14).
Participants in weight loss trials receive extensive attention and training during the trial; it is unclear whether persons in the community who are not part of a weight loss trial and who undergo similar amounts of weight reduction also experience similar changes in pain and function. It is also unknown whether persons who gain weight actually experience worse pain and function and whether this pain and functional loss is proportional to the amount of weight gained. Knowing whether persons in the community report proportionate reductions (or increases) in pain and function following changes in body weight would equip clinicians with additional evidence-based information to aid in the management of knee OA. The purpose of our longitudinal cohort study was to determine if a dose-response relationship exists between the extent of weight changes (including both weight reduction and weight gain) and the extent of changes in self-reported function-related pain and functional status.
A dose-response relationship was found between changes in body weight and subsequent self-reported pain and functional status over an approximately 3-year period.
The threshold for statistically significant changes in pain and function appears to be a ≥10% weight gain or weight reduction.
This study indicates that body weight changes are associated with changes in pain and functional status in a dose-response fashion.
We analyzed data from 2 public use data sets. The Osteoarthritis Initiative (OAI) is a publicly and privately funded prospective longitudinal cohort study with 4 years of followup. A primary objective of the OAI was to develop diverse cohorts of persons for the study of the natural history, risk factors, onset, and progression of tibiofemoral knee OA. The Multicenter Osteoarthritis (MOST) study is also a publically funded prospective longitudinal cohort study. The overall aims of the MOST study were to identify novel and modifiable biomechanical factors, bone and joint structural factors, and nutritional factors that affect the occurrence and progression of knee OA. All centers in both studies required all subjects to read and sign institutional review board–approved consent forms prior to participation.
In the OAI, subjects between the ages of 45 and 79 years with or at high risk for developing knee OA were recruited from communities in and around 4 clinical sites: the University of Maryland School of Medicine in Baltimore, Maryland, the Ohio State University in Columbus, Ohio, the University of Pittsburgh in Pittsburgh, Pennsylvania, and the Memorial Hospital of Rhode Island in Pawtucket, Rhode Island. Persons recruited for the MOST study also had or were at high risk for developing knee OA and were ages 50–79 years. The MOST subjects were recruited from communities in and around 2 clinical sites: the University of Iowa in Iowa City, Iowa, and the University of Alabama, Birmingham in Birmingham, Alabama. Details of the study populations from both cohorts have been described in detail elsewhere (15, 16).
Persons from both the OAI and MOST study had to have the following features to be recruited for our study: 1) radiographic tibiofemoral knee OA, defined as definite osteophytes (Osteoarthritis Research Society International atlas grade 1–3  in the OAI or Kellgren/Lawrence grade 2 or higher  in the MOST study) as measured on a standardized fixed-flexion radiograph (19, 20), 2) a WOMAC pain scale score of 4 or higher, 3) a WOMAC physical function score of 9 or higher, and 4) no knee replacement surgery during the followup period. We intended to study a sample of persons who had radiologically confirmed knee OA and function-limiting pain. Minimal detectable clinical improvement estimates for WOMAC physical function scores generally range from 7–9 points, while for the WOMAC pain scale, estimates range from 2–4 points (21–26). We chose the more conservative change score criteria of 9 points for the WOMAC physical function scale (26) and 4 points for the WOMAC pain scale (25) to reduce the chances of falsely categorizing a person as changed when in fact they had not changed. Because we were interested in determining the effects associated with differing amounts of weight change, we wanted a sample with WOMAC scores that were substantial enough to allow for the detection of change at the individual person level to allow for interpretation. We excluded persons who underwent knee arthroplasty because the surgery would have likely resulted in dramatic changes in WOMAC and performance-based measures (27, 28), and we were interested in weight loss effects in nonsurgical cases. The complete protocol for the OAI can be viewed online (http://www.oai.ucsf.edu/datarelease/docs/StudyDesignProtocol.pdf), and details of the MOST study are also available online (http://most.ucsf.edu/about.asp).
The baseline variables were the knee radiographic data, age, sex, comorbidity status measured on a continuous scale (29), body weight (in kg), race (dichotomized as African American or other), presence of frequent low back pain, depression status using a validated cut score of 16 or higher indicating likely clinical depression on the Center for Epidemiologic Studies Depression Scale (30, 31), a dichotomized variable indicating whether the person was not working (at least in part) for health reasons, education level (less than a high school diploma, high school diploma, or at least some college), and current smoking status. We also included a variable that was coded to indicate the presence of unilateral or bilateral knee OA. We used the clinical data sets 0.2.2 and 5.2.1 from the OAI web site (http://oai.epi-ucsf.org/datarelease/About.asp). All data for the MOST study are publically available.
The primary predictor variable was changes in body weight from baseline to followup. We chose to use 5 categories of body weight changes: ≥10% body weight reduction, 5–9.9% body weight reduction, 4.9% body weight reduction to 4.9% body weight gain, 5–9.9% body weight gain, and ≥10% body weight gain. We chose these categories to reflect the weight changes because they are similar to the categories of weight changes generally described in the literature (6, 32) and in recommendations from government agencies (13, 14). In addition, these categories allow for the assessment of the effects of proportionally similar changes in body weight on pain and function.
The outcome measures were the change scores calculated by subtracting the followup scores from baseline scores. The followup measures were obtained during the 3-year visit for the OAI and at the 30-month followup visit for the MOST study. The outcome variables were identical for both data sets and consisted of the WOMAC pain scale change scores and the WOMAC physical function scale change scores. Both outcome measures have demonstrated high levels of reliability and validity (26, 33–36). WOMAC Likert version 3.1 physical function scores range from 0–68, with higher scores indicating worse function. WOMAC pain scores range from 0–20, with higher scores indicating greater function-related pain (37).
We used chi-square tests for categorical baseline variables and t-tests for continuous variables to compare persons who had followup weight data and those whose followup weight data were missing. The subsequent analyses were performed for the dependent variables of changes in WOMAC function score and changes in WOMAC pain score. We investigated our primary question by first performing 2 regression analyses that included WOMAC function score and WOMAC pain score as dependent variables and weight changes categorized into 5 groups (≥10% reduction, 5–9.9% reduction, 4.9% reduction to 4.9% gain, 5–9.9% gain, and ≥10% gain) as the independent variable. These analyses tested whether function or pain differed among the weight reduction or gain categories compared to the reference category (4.9% reduction to 4.9% gain). Next, we repeated these analyses adjusting for the following covariates: baseline scores for the dependent variable of interest (i.e., either WOMAC function or pain scores), sex (2 levels: female, male) (38), depression (2 levels: depressed, not depressed) (39), and number of comorbidities (40). Our regression model was as follows: changes in WOMAC function or pain = constant + b1(≥10% weight reduction) + b2(5–9.9% weight reduction) + b3(5–9.9% weight gain) + b4(≥10% weight gain) + b5(baseline function or pain) + b6(number of comorbidities) + b7(sex) + b8(depression).
We applied the following dummy variable coding scheme: 1 if the weight category applies, 0 if otherwise; 1 if female, 0 if male; and 1 if depressed, 0 if not depressed. Applying this coding scheme, the reference weight category was the 4.9% weight reduction to 4.9% weight gain category (i.e., b1 = b2 = b3 = b4 = 0). We performed a trend analysis to assess whether the results were consistent with a dose-response relationship. Specifically, we examined the extent to which linear, quadratic, cubic, and quartic trends were evident.
Prior to initiating the analyses, we examined the distributional properties of the variables and checked for heterogeneity of dependent variable variances among the 5 weight change categories. For all analyses, we applied 2-tailed tests and an effect was considered statistically significant if a P value was less than 0.05. Analyses were conducted using Stata, version 10.1.
The characteristics of the sample are shown in Table 1. There were some demographic differences between the 2 data sets. For example, persons from the OAI data set were generally younger, had a higher education level, and weighed less than the persons in the MOST study data set. In addition, there was evidence of selective loss to followup when comparing persons who had followup weight data (n = 1,410) to persons whose followup weight data were missing (n = 375). For example, persons with missing followup weight measures had a lower level of education, were more frequently African American, and had higher levels of pain and worse function. The distribution of persons in each of the 5 weight change categories were n = 82 in the ≥10% weight reduction group, n = 176 in the 5–9.9% weight reduction group, n = 953 in the 4.9% weight reduction to 4.9% weight gain group, n = 148 in the 5–9.9% weight gain group, and n = 51 in the ≥10% weight gain group.
|OAI baseline (n = 976)||MOST baseline (n = 809)||Combined baseline (n = 1,785)||Complete weight data at baseline and followup (n = 1,410)||Missing weight data at followup (n = 375)||Comparing complete and missing data, χ2 or t-test (P)|
|Age, years||61.72 ± 9.25||63.74 ± 8.09||62.63 ± 8.79||62.73 ± 8.62||62.26 ± 9.38||0.94 (0.35)|
|Women, %||62.2||64.5||63.2||62.5||66.1||1.70 (0.19)|
|African American, %||33.5||22.9||28.7||25.5||40.8||34.07 (< 0.001)|
|Marital status, %||16.4 (0.002)|
|Education, %||15.0 (0.001)|
|Less than high school diploma||6.7||7.8||7.2||6.2||11.1|
|High school diploma||18.9||30.2||24.0||23.2||23.7|
|At least some college||74.4||62.1||68.8||70.6||61.7|
|Comorbidity||0.55 ± 0.95||0.64 ± 1.01||0.59 ± 0.98||0.57 ± 0.95||0.67 ± 1.07||1.80 (0.07)|
|Weight, kg||86.63 ± 16.50||93.55 ± 20.98||89.80 ± 19.98||89.62 ± 18.6||90.32 ± 20.25||0.63 (0.53)|
|Current smoker, %||8.6||13.1||9.8||9.0||12.6||3.58 (0.06)|
|CES-D, % depressed||17.9||19.2||18.5||17.7||21.6||3.03 (0.08)|
|WOMAC pain score||7.96 ± 3.20||8.15 ± 3.11||8.05 ± 3.17||7.83 ± 3.06||8.85 ± 3.42||5.57 (< 0.001)|
|WOMAC physical function score||25.23 ± 10.76||26.73 ± 10.29||25.91 ± 10.57||25.12 ± 10.14||28.88 ± 11.63||6.17 (< 0.001)|
The distributions of the dependent variables approximated a normal distribution, and for each dependent variable, the variances among the 5 weight categories did not differ statistically. Table 2 shows a descriptive summary of the WOMAC physical function and pain change scores for the 5 weight change categories as well as the percentage of patients whose change scores met or exceeded the minimum clinically important thresholds of 9 WOMAC physical function points (26) or 4 WOMAC pain points (25). The unadjusted analyses revealed statistically significant differences among the weight change categories for both WOMAC function (F[4,1381] = 4.72, P = 0.001) and pain (F[4,1401] = 2.50, P = 0.041). Table 3 shows the difference between the weight change categories and the reference category (4.9% weight reduction to 4.9% weight gain) in the WOMAC function and pain points. Unadjusted and adjusted coefficients are also shown in Table 3. The results show that for both the unadjusted and adjusted analyses, only the weight change categories ≥10% weight reduction and ≥10% weight gain differed from the reference category. The dose-response relationship for weight changes and the WOMAC physical function scale is shown in Figure 1, while Figure 2 shows the dose-response relationship for the WOMAC pain scale. The trend analyses identified statistically significant linear and cubic trends for both the WOMAC physical function (linear: F[1,1362] = 16.47, P < 0.001 and cubic: F[1,1362] = 12.11, P < 0.001) and pain (linear: F[1,1381] = 8.77, P < 0.003 and cubic: F[1,1381] = 13.79, P < 0.001) measures. The linear trend explains the extent to which there was a steady decline in the scores of the outcome measures across the weight change categories, while the cubic trend supports that there was a flattening or similarity in the scores of the outcome measures for the middle 3 weight change categories.
|Weight change group||WOMAC physical function, mean ± SD change||WOMAC physical function change score of ≥9 points, no./total (%)||WOMAC pain, mean ± SD change||WOMAC pain change score of ≥4 points, no./total (%)|
|≥10% reduction||7.50 ± 13.24||37/82 (45.1)||2.05 ± 4.60||31/82 (37.8)|
|5–9.9% reduction||3.34 ± 12.62||50/171 (29.2)||0.99 ± 4.34||33/171 (19.3)|
|4.9% reduction to 4.9% gain||2.78 ± 11.82||269/940 (28.6)||1.09 ± 3.86||233/951 (24.5)|
|5–9.9% gain||3.23 ± 12.34||41/145 (28.3)||1.40 ± 3.99||42/148 (28.4)|
|≥10% gain||−1.67 ± 13.6||10/48 (20.8)||−0.06 ± 3.93||8/51 (15.7)|
|Unadjusted analysis||Adjusted analysis|
|WOMAC function regression coefficient (95% CI), P||WOMAC pain regression coefficient (95% CI), P||WOMAC function regression coefficient (95% CI), P||WOMAC pain regression coefficient (95% CI), P|
|Weight change category|
|≥10% reduction||4.71 (1.97, 7.45), 0.001||0.96 (0.06, 1.86), 0.036||4.07 (1.49, 6.65), 0.002||0.90 (0.06, 1.74), 0.035|
|5–9.9% reduction||0.55 (−1.42, 2.53), 0.582||−0.10 (−0.74, 0.54), 0.760||0.01 (−1.87, 1.89), 0.991||−0.26 (−0.87, 0.35), 0.402|
|4.9% reduction to 4.9% gain||Ref.||Ref.||Ref.||Ref.|
|5–9.9% gain||0.44 (−1.68, 2.57), 0.682||0.31 (−0.37, 1.00), 0.371||1.08 (−0.91, 3.07), 0.288||0.50 (−0.14, 1.15), 0.128|
|≥10% gain||−4.45 (−7.98, −0.93), 0.013||−1.15 (−2.27, −0.03), 0.045||−5.36 (−8.74, −2.00), 0.002||−1.56 (−2.62, −0.49), 0.004|
|Baseline dependent||N/A||N/A||0.43 (0.36, 0.49), < 0.001||0.49 (0.42, 0.56), < 0.001|
|Female||N/A||N/A||−1.60 (−2.85, −0.35), 0.012||−0.39 (−0.80, 0.01), 0.058|
|Comorbidity (number)||N/A||N/A||−1.26 (−1.91, −0.61), < 0.001||−0.33 (−0.55, −0.12), 0.002|
|Depressed (yes)||N/A||N/A||−2.02 (−3.73, −0.31), 0.020||−0.96 (−1.50, −0.42), 0.001|
|Constant||2.78 (2.01, 3.56), < 0.001||1.08 (0.83, 1.34), < 0.001||−4.2 (−6.73, −1.68), 0.001||−1.74 (−2.58, −0.92), < 0.001|
Our study shows that there was a dose-response relationship between body weight changes and changes in self-reported pain and functional status. Such a study would at a minimum require a large sample of persons followed up over an extended period of time; well-defined methods for quantifying arthritis, pain, and functional status; a high rate of followup; and a sufficient number of persons who either lost or gained weight during the followup period.
We found that persons who lost ≥10% of their body weight over an approximately 3-year period reported significantly lower function-related pain and improved functional status relative to the reference category. In contrast, persons who gained ≥10% of their baseline body weight had significantly worse function-related pain and function than persons in the reference category. The dose-response relationship between weight changes and pain and functional status changes was highly significant (P < 0.003) both for linear and cubic trends.
We found no other evidence that quantified the potential impact of body weight gain on subsequent pain and functional status. Our study suggests that body weight gains of ≥10% body weight have significant effects particularly on self-reported function, but also on pain. After adjusting for covariates, mean WOMAC physical function scores worsened (increased) by 5.4 points (95% confidence interval [95% CI] 2.0, 8.7 points), while WOMAC pain scores worsened (increased) by 1.6 points (95% CI 0.5, 2.6 points) relative to scores in the reference group. While these differences are approximately half the amount of difference required to infer change at the level of the individual patient (21–26), for group-level changes, these estimates approximate the magnitude of changes reported in successful weight loss (6–8) and knee exercise (41) trials. When interpreting the meaningfulness of these group-level changes, it is necessary to distinguish between an important within-patient change and an important between-group difference. Goldsmith and colleagues have shown that an important within-patient change is substantially greater than an important between-group difference (42). For example, randomized trials of rheumatoid arthritis patients have found an important within-patient change in pain to be 36% of the baseline scores compared to 20% of the baseline scores for a between-group difference. Similarly, Goldsmith et al identified an important within-patient change in disability to be 49% of the baseline scores compared to 16% of the baseline scores for a between-group difference (42). Applying the percentage difference estimates reported by Goldsmith et al to the 9-point WOMAC physical function (26) and 4-point WOMAC pain (25) minimal detectable change estimates used in our study yields important between-group differences of ∼2.2 and ∼2.9 WOMAC pain and function points, respectively. Referring to Table 3, the mean disability change scores of the ≥10% weight reduction and ≥10% weight gain groups differ significantly from the reference group. For pain, all between-group comparisons with the reference group are less than 2.2 points.
Table 2 delineates the percentages of patients who met or exceeded the minimum clinically important thresholds for important within-person change. For example, 45.1% of the patients who lost ≥10% of body weight reported changes that met or exceeded the criterion of 9 WOMAC physical function points, while 29.2% of persons who lost between 5% and 9.9% of body weight met the WOMAC physical function change threshold. Similar estimates were reported for the WOMAC pain scale. These data suggest that weight changes of ≥10% appear to be important thresholds for individual patient pain and function changes.
Most of the studies examining the effects of body weight changes on pain and function have been directed toward weight loss. Trials have consistently shown that a 5% or greater weight loss resulted in significantly reduced pain or improved function. For example, Messier and colleagues examined the efficacy of a diet and exercise intervention on obese persons with symptomatic knee OA (8). The authors found that a mean 5.7% weight loss was associated with significant reductions of 5.7 WOMAC physical function points and 2.2 WOMAC pain points after 18 months. These WOMAC changes are somewhat larger than our estimates for persons with ≥10% loss in body weight (point estimates of 4.1 WOMAC physical function points and 0.9 WOMAC pain points). In another trial in which loss in body weight approached 9% in an experimental group, reductions in WOMAC physical function scores were even greater, with a mean 8.4-point reduction in 6 months (9). Bliddal and colleagues reported a similar magnitude of changes in both WOMAC physical function and pain scores in their 1-year trial of persons who lost a mean of 11% of body weight (4). The baseline WOMAC pain and physical function scores in these previous trials were very similar to our sample, suggesting that our sample was similar although with a lower body mass index than these previous trials.
For persons in our study with comparable weight reduction (between a 5% and 9.9% weight loss) to that reported in 2 successful trials (8, 9) and a systematic review (6), changes in WOMAC pain and physical function scores were essentially nonexistent. We suspect this may be related to the fact that persons in our study were not enrolled in a weight loss study and did not receive the additional training and attention compared to persons in a weight loss trial. The effects of weight loss on pain and function, when assessed in multiple sites in the context of a cohort study like ours with no focus on weight changes per se, may dilute the effects as compared to those seen in trials. It is also possible that the longer followup period in our study influenced the effect of weight loss on pain and function. Even with these differences in study design and length, weight loss, when appreciable (≥10% of body weight), was shown to have therapeutic effects.
The National Institutes of Health (NIH) published clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults (14). The recommendation for the extent of weight loss is the following: “the initial goal of weight loss therapy should be to reduce body weight by approximately 10 percent from baseline.” While the NIH guidelines are not designed specifically for persons with knee OA, our findings support the NIH recommendation for weight loss when applied to OA-related pain reduction and functional improvement.
Our study has some notable strengths, but also some important limitations. First, despite our large sample size, the loss to followup was substantial in that 21% of the sample did not have followup weight data. Those lost to followup were generally more symptomatic, tended to be African American, and were less likely to report being married. A selective loss to followup, while not uncommon in large cohort studies (43), likely impacted our findings. We suspect that this loss likely diluted the effects of weight changes, given that those lost to followup were more symptomatic. Future research should focus on enhanced methods of recruitment and retention for these at-risk populations. In addition, persons in the OAI and MOST study data sets met all of the inclusion criteria, but they also differed in several ways. We chose not to adjust for these differences. In total, the combined data set is more heterogeneous than either the OAI or MOST study data sets in isolation and therefore better reflects variation in the types of patients with knee OA seen in clinical practice. The sample sizes for the ≥10% weight gain and ≥10% weight loss groups (n = 51 and n = 82, respectively) were fairly small relative to the other weight change categories. Despite these relatively small samples, our findings were consistent and statistically robust. Finally, our study design was descriptive in nature, and we cannot determine whether the weight loss (or gain) caused predictable changes in pain and functional status or vice versa.
The results of our study have the potential to impact clinical practice. Clinicians should encourage patients who are overweight to lose weight, and the target magnitude of weight loss, based on our study and the NIH recommendations (14), should be 10% or more of body weight. Guidance can also be provided regarding weight gain and the potential impact of future weight gain on subsequent pain and functional status. Patients can benefit by knowing that the dose of changes in body weight is important and is related to pain and functional status.
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 published. Dr. Riddle 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. Riddle, Stratford.
Acquisition of data. Riddle.
Analysis and interpretation of data. Riddle, Stratford.
Merck, Novartis, Glaxosmithkline, and Pfizer had no role in the study design or in the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by Merck, Novartis, Glaxosmithkline, and Pfizer.