Obesity and Joint Health

^aAssessment of hip OA was only carried out for cases. ^bFigures in parenthesis are the confidence intervals. 1, cases; 2 and 3, control group.

NHANES, National Health and Nutrition Examination Survey; JSN, joint space narrowing; JSW, joint space width; i.v., intravenous; HRT, hormone replacement therapy.

Results of the studies included

Nine of the 12 studies showed exact data of the outcomes; the other three described the outcome in global terms only [26, 28, 29].

Seven studies showed a positive association between obesity and hip OA [odds ratio (OR) ≥1.25] [18, 20, 21, 23–25, 27]; in five studies the association was statistically significant [18, 20, 23, 24, 27], indicating that subjects with a higher BMI (approximately >25) have an increased risk of developing hip OA. Four of these seven studies had a case–control design and the other three a cross‐sectional design. Three studies showed a weak positive relationship [relative risk (RR) or OR 1–1.25] [4, 19, 26]; one of these studies had a prospective cohort design.

There were no studies reporting a negative association. There were four studies reporting outcomes on the association between moderate obesity (BMI >25) and hip OA, but also between more severe obesity (BMI >27) and hip OA [18, 20, 21, 25]. Three of these studies showed a clear dose–response relationship [18, 20, 25]. Three studies also investigated the relationship of the BMI and the presence of bilateral disease [4, 18, 20]. The outcomes show that the strength of the relationship was similar to that found for unilateral hip OA.

In contrast with the nine studies which used clinical information (e.g. physical complaints, need for a THR) for the diagnosis of hip OA, the three studies using only an X‐ray as evidence for hip OA [4, 26, 28] reported no association or only a very weak positive association with hip OA.

Methodological quality assessment

The two reviewers scored 404 items and agreed on 378 items (94%, κ = 0.87). The 26 disagreements were resolved in a single consensus meeting. Table 3shows the studies in order of their methodological quality score, subdivided into the different types of study design (i.e. cohort, case–control and cross‐sectional). The scores ranged from 15 to 77% of the maximum obtainable score for each study design. The mean quality score was 57%.

The participation rates of the studied populations (Table 1, items 3, 4) were very low; the criteria were fulfilled by three of the four case–control studies and the cohort study but by only one out of the seven cross‐sectional studies. In contrast with the assessment of obesity, for which there was almost always an identical or comparable assessment within the different groups of the population (Table 1, item 7), the assessment of hip OA (item 9) diverged substantially. In studies with a cohort as study base (longitudinal and cross‐sectional data), there was always an identical assessment of hip OA. In case–control studies (with longitudinal and cross‐sectional data), however, the assessment of hip OA in six of the seven studies was carried out in the patients but not (or differently) in the controls [21–23, 25, 27, 29].

One of the main concerns about the internal validity of observational studies is that of confounding [31]. When there is confounding, the observed effect may occur due to a combination of several determinants that are prevalent in the study population. In hip OA two potential confounders are age and gender, and there should be correction for at least these determinants in the individual studies. Only one study did not meet this criterion [29]; all the others did, by means of matching, restriction or adjustment in the analysis.

No correlation was found between the quality score and the study outcome (Fig. 2) (Pearson correlation = −0.04, P = 0.9).

Three of the 12 studies could not be plotted due to a lack of precise data on the OR or RR. Of these, van Saase et al. [28] achieved a quality score of 67% and reported no association between obesity and hip OA. Hartz et al. [26] scored 56% and reported a weak association, whereas Saville and Dickson [29] achieved a score of 17% and showed no association between obesity and hip OA.

Quality scores of the studies

Each item was scored 1 when it met the specified criteria listed in Table 1. If it did not meet the criteria or was not described at all, a score of 0 was assigned. Positive scores were summed to give an overall internal validity score.

^aInformativity item; not included in the analysis.

na, not applicable.

Best evidence synthesis

According to Chalmers and colleagues [15], the cut‐off point for high‐quality studies was set at the mean of the quality scores, so a study was considered to be of high quality if the methodological quality score was ≥60%.

There were five longitudinal studies, and four of them reached the level of high quality. The high‐quality cohort study reported an OR of 1.03 [95% confidence interval (CI) 0.41–2.60], whereas the three high‐quality case–control studies reported an association of 2.3 (95% CI 1.2–4.4), 3.4 (95% CI 0.4–25.6) and 2.1 (95% CI 0.9–4.6). This means that there is moderate evidence for a positive association between obesity and hip OA, with an OR of approximately 2.

In the subgroup of studies reporting the association between bilateral hip OA and obesity, only three cross‐sectional studies were found, of which two reached the level of high quality [18, 20]. This means that, with the best‐evidence synthesis, we can conclude that there is only limited evidence for a positive association between obesity and the occurrence of bilateral hip OA. The strength of the association is comparable to the strength found for unilateral hip OA.

For the subgroup of articles assessing hip OA on clinical information (e.g. physical complaints, need for a THR), six [18–22, 25] of the nine articles reached the level of high quality. One had a cohort design [19], three a case–control design [21, 22, 25] and two a cross‐sectional design [18, 20]. The reported ORs were 1.03, 1.4, 2.3, 2.1, 1.6 and 2.0. This means that there is moderate evidence for a positive relationship between clinically assessed hip OA and obesity.

All three studies that assessed hip OA with radiological parameters had a cross‐sectional design [4, 26, 28]. The reported outcomes were ‘absence of an association’, ‘a weak association’ and an OR of 1.02. One of them reached the level of high quality [32], so there is limited evidence for no association in the subgroup of radiologically assessed hip OA.

Discussion

In this systematic review, we summarized the evidence available in the literature on the influence of obesity on the development of OA of the hip. On the basis of this evidence, we may conclude that there is moderate evidence for a positive association between hip OA and obesity. For the subcategory of bilateral hip OA, only limited evidence was found. The strength of association found in this subgroup was comparable to the strength found in the overall conclusion.

For a subcategory of articles in which hip OA was assessed clinically, moderate evidence was also found for a positive association, in contrast to radiologically assessed hip OA, for which limited evidence for no association was found. This review may, however, suffer from several restrictions.

Identification and selection of the literature

Although we put much effort into identifying the relevant articles, our literature search might have had some limitations. Some relevant articles may have been missed because they used other key words or had unclear abstracts, and not all published articles are indexed in databases. Besides, we excluded articles written in languages other than English, Dutch, German, French, Danish, Norwegian and Swedish.

The presumed publication bias found in our results (Fig. 1) can be explained by a selection of studies that were published for their ‘exciting’ positive results that satisfy current dogma. When assessing the funnel plot, we have to keep in mind that three of the 11 studies could not be included in the analysis due to insufficient data to deduce a point estimate [26, 28, 29].

Unfortunately, only one prospective study of the association between obesity and hip OA could be found. This design is known to be the most valid type of observational studies. Having more data from this type of design would have allowed us to provide a more valid and precise conclusion. The decision to include cross‐sectional studies is based on the results of several studies reporting a similarity between current BMI and historical BMI in patients with hip OA [25, 33, 34]. Our data support these findings. We acknowledge, however, that the strength of the evidence of cross‐sectional studies is much lower than the evidence provided by prospective cohort studies and, to a lesser extent, case–control studies.

Quality assessment and best‐evidence synthesis

The quality assessment was challenging because no tested and validated criteria lists have been published for observational studies in the field of OA. Also, limited data were found on performing a best‐evidence synthesis with observational studies. Therefore we have presented them in a reproducible manner, and the criteria we used were relatively strict.

A very notable item in the quality assessment was the difference in the assessment of hip OA between the patients and controls in the case–control studies (either longitudinal or cross‐sectional). Most studies took as cases all patients with hip OA and as controls a sample of the underlying population, but without screening them for hip OA. This means that some controls might have had hip OA, so these studies would be biased towards a null result.

There have been suggestions that controls do not need to be evaluated for the occurrence of the disease, but rather must represent the base population, which will include such cases [35]. Considering the fact that there are still contradictory opinions on this subject, we prefer the classical point of view, namely that controls must be free of the disease under study.

The most interesting finding in our study is the difference found between clinically assessed and radiologically assessed hip OA. Whereas the former group of studies shows that obese people suffer more from hip OA, the latter shows that there is no difference in the presence of hip OA, between obese and non‐obese people. This may suggest that obese people suffer more from the same radiological degree of hip OA than non‐obese people do. Thus, obese patients may have more hip complaints at an equal radiological stage, due to the weight on the hip joint, and therefore qualify earlier for THR than non‐obese patients. Unfortunately, there are only a few studies reporting on weight loss in relation to knee OA, and none in relation to hip OA, to support this explanation [36, 37].

In view of the moderate evidence for an association, knowledge of the origin of the relationship between hip OA and obesity is important. Future research, especially well‐designed prospective cohort studies with adequate follow‐up time, will not only strengthen the conclusion of this review but might also throw light on the cause of this relationship.

Conclusion

This systematic review shows that in the literature, there is moderate evidence for a positive influence of obesity on the development of hip OA.

This study was supported by a grant from the Dutch Arthritis Association.