SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

Objective

To examine longitudinal patterns in body mass index (BMI) over 14 years and its association with knee pain in the Chingford Study.

Methods

We studied a total of 594 women with BMI data from clinic visits at years (Y) 1, 5, 10, and 15. Knee pain at Y15 was assessed by questionnaire. Associations between BMI over 14 years and knee pain at Y15 were examined using logistic regression.

Results

BMI significantly increased from Y1 to Y15 (P < 0.0005) with medians (interquartile ranges) of 24.5 kg/m2 (22.5–27.2 kg/m2) and 26.5 kg/m2 (23.9–30.1 kg/m2), respectively. At Y15, 45.1% of subjects had knee pain. A greater BMI at Y1 (odds ratio [OR] 1.34, 95% confidence interval [95% CI] 1.05–1.69), at Y15 (OR 1.34, 95% CI 1.10–1.61), and change in BMI over 15 years (OR 1.40, 95% CI 1.00–1.93) were significant predictors of knee pain at Y15 (P < 0.05). BMI change was associated with bilateral (OR 1.61, 95% CI 1.05–1.76, P = 0.024) but not unilateral knee pain (OR 1.22, 95% CI 0.73–1.76, P = 0.298). The association between BMI change and knee pain was independent of radiographic knee osteoarthritis (OA). The strength of association between BMI and knee pain at Y15 was similar during followup measurements.

Conclusion

Over 14 years, a higher BMI predicts knee pain at Y15 in women, independently of radiographic knee OA. When adjusted, the association was significant in bilateral, not unilateral, knee pain, suggesting alternative pathologic mechanisms may exist. The longitudinal effect of BMI on knee pain at Y15 is equally important at any time point, which may assist reducing the population burden of knee pain.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

Osteoarthritis (OA) is the most common form of arthritis in Western populations. The knee is the main joint affected, with knee pain being the most common and often most disabling symptom in the general population (1). There are more than 6 million people in the UK experiencing knee pain (2).

It is known that there is a global rising obesity epidemic and being overweight/obese is the strongest and most established risk factor for knee OA (3–13). Obesity already costs the UK economy more than £3 billion per year; this cost is expected to increase to £45.5 billion by the year 2050 (2), making this a potentially significant and costly UK health problem.

The question of whether obesity causes OA, or whether having OA leads to becoming obese, is an important matter that has been debated over the years and requires longitudinal data to address this issue. Although several cohort studies have demonstrated strong associations between obesity and radiographic knee OA (RKOA) (9–11, 14–22), most of the published cohort study research is either cross-sectional in design (7, 15, 17, 18), uses body weight measurements taken at baseline but not at followup visits (11, 16), or uses self-reported body weight and height measurements (10, 20, 21), which are likely to have a potential for bias. There are few longitudinal cohorts examining the long-term impact of obesity on knee joints or how body weight measured repeatedly throughout life may allow for better prediction of knee pain rather than RKOA. RKOA, measured by the Kellgren/Lawrence (K/L) scale, has a strong correlation with pain (23, 24); however, the relationship between RKOA and knee pain is by no means perfect (25). Discrepancies do exist, e.g., a subject with a K/L grade of 4 reporting no pain whatsoever and a subject with a K/L grade of 1 reporting severe pain. This remains a focus for current research, although few studies have looked at knee pain longitudinally.

Jinks et al (20) investigated overweight/obesity on the onset and progression of knee pain in older adults living in a North Staffordshire, UK community, and in a later study (26) described the predictors of onset and progression of knee pain in adults. Both studies relied on self-reported height and weight measurements with followup limited to 3 years. Holliday et al (27) recently looked at lifetime body mass index (BMI) and other anthropometric measures of obesity and risk of knee/hip OA in the Genetics of Osteoarthritis and Lifestyle case–control study. The authors concluded that being overweight earlier in adult life increased knee OA and hip OA risk, although this conclusion relies heavily on retrospective estimates of lifetime weight and body shape.

The Framingham study is one cohort study to look at body weight longitudinally (22). Felson et al examined the change in BMI in women up to 12 years prior to knee OA symptom onset; however, all BMI calculations were based on baseline height measurements, which are known to change with age (22). Although findings from the Chingford Study (7) support data from the Framingham cohort study (4), i.e., that asymptomatic obese women are at an increased risk of developing symptomatic RKOA (which leans toward a causal effect of obesity), the need to look at this potential longitudinal relationship between body weight and knee pain is of great importance, since obesity is fast becoming an increasingly serious public health problem worldwide (28). There is currently no cure for knee OA; therefore, identification of risk factors that influence symptomatic knee OA is important. This could guide clinicians to tailor lifestyle and nonoperative measures, such as losing weight at specific time points, to delay or prevent the onset of pain and impairment that may lead to problems with activities of daily living, work disability, and joint replacement surgery.

The main aim of this study is to examine whether BMI measured longitudinally in the Chingford Study can predict knee pain at 15 years and, if so, whether the association is stronger at any particular age and is influenced by RKOA.

Significance & Innovations

  • Over a 14-year period, a higher body mass index (BMI) predicts year 15 (Y15) knee pain in women, independently of radiologic knee osteoarthritis (RKOA).

  • When adjusted, this association was only significant in bilateral, not unilateral, knee pain, suggesting alternative pathologic mechanisms may exist.

  • The longitudinal effect of BMI on Y15 knee pain is equally important at any time point over 14 years, irrespective of RKOA status, which may have significant implications for weight management interventions in decreasing the population burden of knee pain.

PATIENTS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

Patient selection and data collection.

This research utilizes the Chingford 1,000 Women Study based in Essex, Outer London. All 1,353 women, ages 45–64 years (translating to an actual age range of 44–67 years during data collection, with 17 patients in the extended range), from a large general practice register of more than 11,000 patients in Chingford, UK, were invited to participate in a longitudinal cohort to study OA and osteoporosis in 1989. Of the 1,353 women invited, 1,003 agreed to participate, 6 died, 66 moved away, and 278 refused or did not respond. This gave a response rate at initial recruitment of 78%. By year 15 (Y15), 99 women had died, 75 moved away, 85 dropped out, 65 declined participation, and 22 were unable to be contacted, resulting in 657 women (66%) still participating after 14 years. The number of women with complete data at all 10 visits over 14 years significantly reduced this sample by one-third to 332 women. However, a total of 594 women had BMI data for visits at Y1, Y5, Y10, Y15, and knee pain data at Y15, and were therefore included in this study and were known as the attendee cohort. The study was approved by the local research ethics committee, and written consent was obtained from each woman.

Height was measured in centimeters (to the nearest 0.1 cm) in a standing position, with shoes removed, using a wall-mounted stadiometer (Leicester Height Measure, Seca). Weight was measured in kilograms (to the nearest 0.1 kg) by electronic scales with shoes removed. BMI was calculated (weight in kilograms divided by the square of height in meters) and stratified for Figures 1 and 2 using the World Health Organization's (WHO) International Classification for BMI: underweight <18.50 kg/m2; normal 18.50–24.99 kg/m2; pre-obese 25.00–29.99 kg/m2; and obese ≥30.00 kg/m2 (29). Further statistical analyses used BMI as a continuous variable.

thumbnail image

Figure 1. Body mass index over time stratified by World Health Organization categories.

Download figure to PowerPoint

thumbnail image

Figure 2. Pattern plots of body mass index (BMI) with 4-letter sequences representing BMI World Health Organization categories over 4 time points (year 1 [Y1], Y5, Y10, and Y15). U = underweight; N = normal; P = pre-obese; O = obese.

Download figure to PowerPoint

A standardized joint symptom questionnaire obtained details of knee pain at Y3 and Y15 using 2 questions: “Have you had any knee pain in either knee in the last month?” and “How many days of pain have you experienced in the last month?” Responses from the first knee pain question were categorized into no pain, unilateral pain, or bilateral knee pain for analysis.

A standardized nurse-administered questionnaire obtained information on demographic variables such as smoking status, alcohol intake, occupation, and physical activity. A total physical activity score was calculated at Y1 by combining the 3 raw physical activity scores relating to walking, occupation, and sport. These activities were each graded 1–4 (where 1 is low and 4 is high), giving a total physical activity score of 3–12.

Standard weight-bearing anteroposterior knee radiographs with legs in full extension were taken by experienced radiographers using the same radiography equipment and protocol at Y1 and Y15 (14, 30, 31). The back of the knee was kept in contact with the cassette. A tube-to-film distance of 100 cm was used, with the beam centered 2.5 cm below the apex of the patella (31). All radiographs were read individually and were blinded to sequence, patient identity, and symptoms. Y1 radiographs were read by 2 observers (DJH and TDS) and Y15 radiographs were read by a single observer (DJH). Inter- and intraobserver reproducibility for these readings have previously been reported with kappas of >0.8 (32). Radiographs were scored according to K/L grades 0–4 (33), using the Atlas of Standard Radiographs (34). An individual was considered positive for RKOA if a K/L grade of ≥2 was present on either knee.

Statistical analyses.

Statistical analyses were restricted to subjects with complete data for BMI at Y1, Y5, Y10, and Y15, and knee pain data at Y15 (n = 594). Descriptive statistics of attendees and nonattendees are shown in Table 1 to determine selection bias. For normally distributed continuous data, the 2-tailed independent t-test was used; the Wilcoxon–Mann-Whitney test was used for non-normal data. For categorical data, the chi-square test was used, providing that all cells had expected frequencies of ≥5, otherwise Fisher's exact test was used.

Table 1. Baseline clinical characteristics for different study subsets*
Clinical characteristicsFull cohort (n = 1,003)Nonattendees, Y15 (n = 409)Attendees, Y15 (n = 594)
  • *

    Values are the percentage (number/total number) unless indicated otherwise. Kellgren/Lawrence (K/L) grade calculated using the maximum grade of left and right knee (for subjects with K/L data for both knees only). Y = year; IQR = interquartile range; BMI = body mass index.

  • P < 0.0005.

  • P < 0.05.

  • §

    P < 0.01.

  • For n = 989.

  • #

    For n = 400.

  • **

    For n = 589.

  • ††

    Earliest record, used as baseline data.

Age, median (IQR) years54 (49–60)57 (51–57)52 (48–58)
Height, mean ± SD meters1.62 ± 0.0591.61 ± 0.0591.62 ± 0.059
Weight, median (IQR) kg65.0 (58.5–73.0)66 (58.7–75)64.4 (58.4–71.8)
BMI, median (IQR) kg/m224.8 (22.6–27.6)25.3 (22.9–28.5)24.5 (22.6–27.2)§
Smoking, current or ex46.16 (463/1,003)53.55 (219/409)41.08 (244/594)
Alcohol, current or ex81.75 (820/1,003)78.73 (322/409)83.84 (498/594)
Physical activity score, median (IQR)7 (6–8)7 (6–8)#7 (6–8)**
Y3 knee pain††   
 No pain71.28 (628/881)67.80 (219/323)73.30 (409/558)
 Unilateral9.42 (83/881)9.60 (31/323)9.32 (52/558)
 Bilateral19.30 (170/881)22.60 (73/323)17.38 (97/558)
Y15 knee pain   
 No pain54.40 (352/647)49.06 (26/53)54.88 (326/594)
 Unilateral23.18 (150/647)20.75 (11/53)23.40 (139/594)
 Bilateral22.41 (145/647)30.19 (16/53)21.72 (129/594)
Y1 K/L grade   
 079.34 (768/968)74.61 (288/386)82.47 (480/582)
 16.10 (59/968)7.77 (30/386)4.98 (29/582)
 29.61 (93/968)11.14 (43/386)8.59 (50/582)
 34.86 (47/968)6.22 (24/386)3.95 (23/582)
 40.10 (1/968)0.26 (1/386)0.00 (0/582)

BMI data were further described by classifying it into WHO categories (U = underweight, N = normal, P = pre-obese, O = obese) and observing the proportions in each category at each year for which data were available (Figure 1). The subjects' WHO classifications at the 4 time points used in this study (Y1, Y5, Y10, and Y15) were further observed with their classifications constituting a sequence or pattern over time. Subjects with identical BMI WHO classification patterns were collected and graphically expressed using bar charts (Figure 2). Those who remained in the same classification at all 4 time points were classified as “stable BMI.” The classification of “increasing BMI” was given to those subjects with an increase only in WHO classification over time. The classification of “decreasing BMI” described the subjects that only decreased in WHO classifications over time, and the classification of “fluctuating BMI” described those subjects with an increase of ≥1 and a decrease of ≥1 in BMI classifications over time. The pattern plots, based on permutations generated using a combinatorics algorithm, are a descriptive tool used to provide an overview of the BMI movements of subjects over the course of 14 years.

Statistical analyses were used to determine associations between Y15 knee pain (binary outcome: yes/no from the pain question, “Have you had any knee pain in either knee in the last month?”) and 5-unit BMI (exposure) at Y1, Y15, and the BMI change between these years (Y15 minus Y1) as shown in Table 2. The crude model was then adjusted for baseline age (Y1) and Y3 pain. Table 2 was also adjusted for smoking, alcohol, and physical activity.

Table 2. Subject-level logistic regression analysis using Y15 knee pain (outcome) and 5-unit BMI (exposure)*
Y15 outcome, exposureUnadjusted modelAdjusted for Y1 ageAdjusted for Y1 age and Y3 pain (yes/no)
OR (95% CI)PKnees, no.OR (95% CI)PKnees, no.OR (95% CI)PKnees, no.
  • *

    Y = year; BMI = body mass index; OR = odds ratio; 95% CI = 95% confidence interval.

  • Earliest record, used as baseline data.

  • P < 0.05.

  • §

    P < 0.01.

  • Only value to change (to OR 1.38, 95% CI 1.01–1.88, P = 0.042) when Table 2 was adjusted for smoking, alcohol, and physical activity. All other models remained unaltered by this adjustment.

Y1 BMI         
 Unilateral pain versus no pain1.34 (1.05–1.76)0.0224651.34 (1.05–1.76)0.0264651.28 (0.95–1.69)0.094439
 Bilateral pain versus no pain1.40 (1.10–1.84)0.008§4551.40 (1.05–1.84)0.0124551.28 (0.95–1.76)0.091430
 Any pain versus no pain1.40 (1.10–1.76)0.002§5941.40 (1.10–1.76)0.003§5941.34 (1.05–1.69)0.019558
Y15 BMI         
 Unilateral pain versus no pain1.28 (1.05–1.61)0.0184651.28 (1.05–1.61)0.0174651.22 (1.00–1.54)0.053439
 Bilateral pain versus no pain1.40 (1.16–1.76)0.002§4551.40 (1.10–1.76)0.002§4551.34 (1.05–1.76)0.010430
 Any pain versus no pain1.34 (1.10–1.61)0.001§5941.34 (1.10–1.61)0.001§5941.34 (1.10–1.61)0.003§558
BMI change (Y15–Y1)         
 Unilateral pain versus no pain1.16 (0.82–1.69)0.3464651.22 (0.86–1.76)0.2834651.22 (0.73–1.76)0.298439
 Bilateral pain versus no pain1.47 (1.00–2.19)0.0504551.54 (1.05–2.29)0.0334551.61 (1.05–1.76)0.024430
 Any pain versus no pain1.34 (0.95–1.76)0.0785941.34 (1.00–1.84)0.0515941.40 (1.00–1.93)0.039558

Logistic regression analyses at the knee level clustered by subject identification using a general estimating equation (GEE) was used to identify the association between Y15 knee pain (binary outcome: yes/no from pain question, “Have you had any knee pain in either knee in the last month?”) and 5-unit BMI at Y1, Y15, and the change in BMI. The crude model was then adjusted for baseline age (Y1), Y3 pain, and Y1 and Y15 K/L grades as shown in Table 3. Table 3 was also adjusted for smoking, alcohol, and physical activity.

Table 3. Knee-level logistic regression analysis clustered by subject identification, with Y15 knee pain (outcome) and 5-unit BMI (exposure)*
Y15 outcome, exposureUnadjusted model (n = 1,188 knees)Adjusted for Y1 age (n = 1,188 knees)Adjusted for Y1 age and Y3 pain (n = 1,116 knees)Adjusted for Y1 age, Y3 pain, and Y1 and Y15 K/L grade (n = 996 knees)
OR (95% CI)POR (95% CI)POR (95% CI)POR (95% CI)P
  • *

    All models remained unchanged when adjusted for smoking, alcohol, and physical activity. Y = year; BMI = body mass index; K/L = Kellgren/Lawrence; OR = odds ratio; 95% CI = 95% confidence interval.

  • Y3 pain; earliest record, used as baseline data.

  • P < 0.01.

  • §

    P < 0.0005.

  • P < 0.05.

Pain, Y1 BMI1.34 (1.10–1.61)0.0031.34 (1.10–1.61)0.0051.34 (1.05–1.61)0.0071.16 (0.90–1.40)0.279
Pain, Y15 BMI1.28 (1.10–1.54)0.0011.28 (1.10–1.54)0.0011.34 (1.10–1.54)< 0.0005§1.22 (1.00–1.47)0.026
Pain, BMI change (Y15–Y1)1.28 (1.00–1.69)0.0481.34 (1.05–1.76)0.0281.40 (1.10–1.93)0.0091.40 (1.05–1.93)0.018

Logistic regression analyses with clustering was also performed at each of the 10 time points: Y1–Y6, Y8–Y10, Y15 with Y15 knee pain (outcome), and 5-unit BMI (exposure) for subjects with complete data at all 10 time points over the 14 years (n = 332), as shown in Figure 3. All statistical analyses were conducted using Stata SE software, version 10 (StataCorp), and Matlab R2009b (The Mathworks).

thumbnail image

Figure 3. Odds ratios (95% confidence intervals) from logistic regression analysis using generalized estimating equation, with year 15 knee pain (outcome) and 5-unit body mass index (BMI; exposure) at different time points. Complete cases of BMI used, n = 664 knees (332 subjects).

Download figure to PowerPoint

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

Table 1 shows the clinical characteristics for the full cohort (n = 1,003), nonattendee cohort (n = 409), and attendee cohort (n = 594). Compared to the nonattendee cohort, attendees are younger with a median age of 52 years (5-year difference; P < 0.0005), less heavy at 64.4 kg (1.6 kg difference; P < 0.05), have a lower percentage of smokers at 41.08% (12.47% difference; P < 0.0005), and a higher percentage of alcohol drinkers of 83.84% (5.11% difference; P < 0.05). Although more attendees (73.30%) were not experiencing knee pain at Y3 (5.5% difference; P = 0.152) and also at Y15 (54.88%, 5.82% difference; P = 0.366) compared to nonattendees, the differences were not significant.

Figure 1 shows the distribution of women grouped by the WHO BMI classifications over the 14 years. The BMI median (interquartile range [IQR]) over 14 years gradually increases from 24.5 kg/m2 (IQR 22.5–27.2 kg/m2) at Y1, to 25.6 kg/m2 (IQR 23.0–28.5 kg/m2) at Y5, to 26.2 kg/m2 (IQR 23.5–29.3 kg/m2) at Y10, and on to 26.5 kg/m2 (IQR 23.9–30.1 kg/m2) at Y15. The number of women in the normal BMI category decreases by nearly 20% with 54.5% (324 out of 594 women) at Y1 to 35.2% (209 out of 594 women) at Y15, whereas the number of women in the obese BMI category more than doubles from 10.9% (65 out of 594 women) at Y1 to 25.8% (153 out of 594 women) by Y15. During this period there was an increase in reported knee pain; 26.7% of women (149 out of 558) experienced knee pain at Y3, and by Y15 this rose to 45.1% of women (268 out of 594).

The descriptive pattern plots in Figure 2 describe the attendee cohort's BMI movements in terms of WHO classifications (U = underweight, N = normal, P = pre-obese, and O = obese) over the 4 time points (Y1, Y5, Y10, and Y15). The BMI patterns were generated using a combinatorics algorithm and were subdivided into the BMI categories of stable, increasing, decreasing, and fluctuating. A total of 46 permutations for longitudinal BMI patterns were identified. Stable BMI patterns were most common, accounting for 48.5% of all patterns. Increasing BMI patterns accounted for 34.3% of all patterns with a normal to pre-obese or a pre-obese to obese increase being most common. Decreasing BMI patterns accounted for just 4.9% with a late decrease from pre-obese to normal being most common. Fluctuating BMI patterns accounted for 12.3% of patterns with fluctuations from normal to pre-obese and vice versa, accounting for over half of fluctuations.

Logistic regression analysis in Table 2 looks at the associations between BMI over the 14 years and knee pain at Y15. The analysis was adjusted for age at Y1 and pain at Y3. In the unadjusted model, Y1 BMI and Y15 BMI are significant predictors of all knee pain (unilateral, bilateral, and any knee pain) at Y15, whereas BMI change is only a significant predictor for bilateral Y15 knee pain. When adjusted for Y1 age, all these predictors remained significant; however, when adjusting for Y1 age and Y3 pain, some of the significance of Y1 BMI (unilateral and bilateral Y15 knee pain) and Y15 BMI (unilateral Y15 knee pain) were attenuated, resulting in Y1 BMI only predicting any knee pain and Y15 BMI predicting bilateral and any Y15 knee pain.

Interestingly, the significance of BMI change in this fully adjusted model increased, predicting bilateral pain and any pain. There was no significant interaction between Y3 pain and BMI for Y15 knee pain, and this remained the case when age was added to the model. When Table 2 was adjusted for smoking, alcohol, and physical activity, the only result affected was in the model adjusted for Y1 age, under the BMI change (Y15–Y1) Y15/any pain versus no pain category; the odds ratio (OR), 95% confidence interval (95% CI), and P value changed from OR 1.34 (95% CI 1.00–1.84), P = 0.051 to OR 1.38 (95% CI 1.01–1.88), P = 0.042. All other models in Table 2 remained unaltered by this adjustment.

A sensitivity analysis using the second knee pain question, “How many days of pain have you experienced in the last month?” (with a ≥15-day cutoff), showed overall similar results to those from the first question, “Have you had any knee pain in either knee in the last month?” used in Table 2.

To see if the predictive effect of BMI is mediated by RKOA, logistic regression analyses at the knee level was performed in Table 3. A GEE model was used to account for clustering of the knee within an individual woman. This technique accounts for the correlation between right and left knees, using each woman as the observation unit and the knees as repeated measurements (35).

The unadjusted model showed Y1 BMI, Y15 BMI, and BMI change to all be significant predictors of Y15 knee pain. When adjusted for Y1 age and Y1 age/Y3 knee pain, all these BMI predictors remained significant (in fact, Y15 BMI and BMI change became slightly more significant on adjusting for Y1 age and Y3 pain). Adjusting for Y1 age, Y3 pain, and Y1 and Y15 K/L grades caused Y1 BMI to no longer significantly predict Y15 knee pain, and although Y15 BMI and BMI change remained significant predictors of Y15 knee pain after this adjustment, their significance level was reduced to P < 0.05.

There were no significant changes to these results when the GEE model was further adjusted (individually or collectively) for smoking, alcohol, or physical activity score (results not shown). Y15 K/L grade and Y15 knee pain were highly associated (P < 0.005), and BMI and K/L grade were highly correlated (P < 0.0005 by Spearman's test).

The longitudinal logistic regression analyses in Figure 3 show that the ORs at all 10 time points (Y1–Y6, Y8–Y10, and Y15) are significant and the 95% CIs overlap across all time points, suggesting there is no one particular time point that predicts BMI for knee pain at Y15 better than another. Similar results were produced when the model was adjusted for baseline age (albeit with slightly lower ORs across all time points). Finally, a sensitivity analysis to investigate the influence of 21 total knee replacements by Y15 showed virtually no change in any of the results.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

We have examined longitudinal patterns in BMI over 14 years and its association with knee pain using a well-known community-based prospective cohort of women based in Chingford, UK. Our results show a marked increase in BMI over time, with a reduction of women remaining in the normal BMI category and a substantial increase of women moving into the obese BMI category. This change in BMI is shown to predict knee pain at Y15, bilateral knee pain more than unilateral knee pain. Our results also show there appears to be no particular time point when a change in BMI is more relevant to developing knee pain at Y15.

The increase seen in BMI over time is expected and is consistent with other longitudinal cohorts (9, 14, 22). There are very few longitudinal studies with true physical measurements for height and weight data collected at so many time points as is found in the Chingford cohort. This is a real strength and makes it an extremely valuable data set. Most other cohorts have looked at height and weight measurements at baseline only (7, 11, 15–18), and some rely on self-reported figures, which have a great potential for recall bias thereby calling their reliability into question (10, 20, 21).

The Figure 2 pattern plots described the BMI movements of the attendee cohort in terms of WHO BMI classifications over the 4 time points. This was a useful technique to quantify the movements and extract descriptive statistics. Unfortunately, there was insufficient statistical power to study knee pain risk in the fluctuating and/or decreasing BMI groups; therefore, a continuous BMI change variable was used.

We have shown that the change in BMI can predict knee pain at Y15, particularly bilateral knee pain over unilateral knee pain. There are few other studies examining BMI and bilateral versus unilateral knee pain to which we can compare. In 2006, Jinks et al reported that being overweight (i.e., BMI ≥25 kg/m2) was a predictor of severe knee pain onset at 3 years in an adult North Staffordshire, UK cohort, who were not experiencing knee pain at baseline, and that obesity (BMI ≥30 kg/m2) was also a strong predictor of progression of nonsevere knee pain to severe knee pain at 3 years (20). However, this study relies on self-reported height and weight measurements taken at baseline and followup, knee radiographs were not taken, followup was limited to 3 years, the analysis did not take the effect of any weight change into consideration, and data on laterality of knee pain were collected but unfortunately not reported.

Previous studies from the Chingford cohort have reported nearly an 18-fold increased risk in bilateral knee OA disease for women in the top BMI tertile (7), and more than one-third of middle-aged women with unilateral knee OA will progress to bilateral knee OA disease within 2 years with obesity being a strong and important risk factor for this occurrence (14). Both studies defined knee OA radiographically using a K/L grade of ≥2. The fact that our results showed bilateral Y15 knee pain achieving significant ORs in the fully-adjusted model for Y15 BMI and change in BMI, but not for Y1 BMI, compared to unilateral Y15 knee pain not achieving any significant ORs for either Y1 BMI, Y15 BMI, or BMI change, could suggest a difference in pathogenesis between bilateral and unilateral knee pain. This may support the theory that, similar to hand OA, bilateral knee OA might be linked to an important systemic metabolic component and not due only to a mechanical loading component (8, 13, 36, 37). One of many possible hypotheses could be the role of obesity-related inflammatory processes via adipocytokines, such as the polypeptide leptin, which regulates food intake and energy expenditure at the hypothalamic level (38–40).

Significant associations between Y15 knee pain and BMI are shown in Table 3. However, when adjusting for K/L grade, the relationship between Y1 BMI and Y15 knee pain becomes nonsignificant, suggesting that K/L grade is either acting as a confounder or on the causal pathway. It is possible that K/L grade may be causing knee pain through RKOA, whereas Y15 BMI and change in BMI may have a direct effect on the Y15 pain threshold. This may also be through the role of obesity-related inflammatory processes via adipocytokines (13), or it may simply be that heavier subjects experience greater pain. In 2008, Heim et al (41) showed a higher prevalence of overall body pain among obese men and women compared to their normal-weight peers, and that obese adults are at increased odds of developing pain. In 2003, cross-sectional studies by Andersen et al (42) showed greater knee pain prevalence with increased levels of BMI, and in 2006 Adamson et al (43) also showed a positive relationship between obesity and knee pain.

To our knowledge, the longitudinal effect of BMI on knee pain over a long period has not yet been reported. Most previous longitudinal data are on structural RKOA (4, 6, 16, 19), and we are showing the longitudinal effect of BMI on knee pain, which is novel. Figure 3 shows that the course of the ORs over the 14 years remains significant, and more or less constant over time. This indicates that the longitudinal effect of BMI over 14 years on knee pain developed at Y15 is equally important at any time point, irrespective of RKOA status, and suggests that weight loss at any point in time over 14 years may be useful in decreasing the population burden of knee pain. This may have important implications for the role of weight management interventions in people with knee pain.

The lack of baseline knee pain data is a limitation to this study. It is unfortunate that these data were not collected at the baseline visit; however, we are fortunate to have these data available from Y3 onward so comparisons with the earlier followup visits can be made. The presence and duration of knee pain is available; unfortunately, we do not have any measurement of knee pain severity, which would have been a useful variable to include in our analysis.

Genetics or OA family history and previous knee injury are other potential confounder variables that may lead to knee pain. Unfortunately, we do not have these validated confounders for this cohort at this time, which limits our analysis.

The women in the Chingford study are a well-described, predominantly white cohort who are representative of the general UK female population in terms of height, weight, and rates of hysterectomy, but with a lower percentage of current smokers (7, 44). Due to the original study design, the results of this study are restricted to white women, thereby limiting generalizability to other types of populations.

Subjects lost to followup are a major limitation of all long-term cohort studies, and the Chingford study is no exception. Unfortunately, the 41% (409 out of 1,003 women) dropout rate is an unavoidable limitation; however, to have 594 women out of the original 1,003 still attending after 15 years is an excellent attendance rate considering the 15-year age increase. The fact that the same study coordinator who started on this project continues to work with these women is a real strength in terms of maintaining excellent rapport and attendance rates.

In conclusion, this study examines the longitudinal relationship between BMI and knee pain in women over a 14-year period, showing that a higher BMI predicts knee pain at Y15. This association is greater with bilateral over unilateral knee pain, suggesting alternative pathologic mechanisms may exist. Longitudinal BMI patterns are described and show that BMI, irrespective of the time point and RKOA status over 14 years, predicts knee pain at Y15. This may have significant implications for weight management interventions in decreasing the population burden of knee pain.

AUTHOR CONTRIBUTIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

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. Ms Goulston 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. Goulston, Spector, Arden.

Acquisition of data. Goulston, Soni, White, Hart, Spector, Arden.

Analysis and interpretation of data. Goulston, Kiran, Javaid, Arden.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES

Thanks are extended to the 1,003 Chingford women for participating, Maxine Daniels for assistance with data collection, and to the rest of the Chingford research team.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. PATIENTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. Acknowledgements
  9. REFERENCES