Background

The hip is the third joint most commonly affected by osteoarthritis (OA) [1]. No therapeutic cure exists for hip OA. Therefore, predicting which patients with hip OA are more likely to progress in their disease is of special interest, particularly if these predictive factors are potentially modifiable.

In 2002, Lievense et al. published a systematic review in which they identified several predictive factors for the progression of hip OA [2]. They used a best-evidence synthesis to draw conclusions about the available evidence per factor. Strong evidence was found for more rapid progression in patients with a superior or superolateral migration of the femoral head or an atrophic bone response. Conversely, strong evidence was found for no association between progression of hip OA and obesity. In 2009, Wright et al. also reviewed the known prognostic factors and their quality of evidence [3]. They concluded that only a few factors are strongly associated with the progression of hip OA, i.e., age, joint space width, migration of the femoral head, femoral osteophytes, bony sclerosis, Kellgren and Lawrence (K-L) grade 3, hip pain at baseline, and a Lequesne index score > 10. In that review, acetabular osteophytes showed no association with progression. Furthermore, de Rooij et al. studied the factors predicting the course of pain and function. They found strong evidence that higher comorbidity count and lower vitality predict a worsening of physical function [4]. Although all reviews described additional predictive factors, the evidence for these factors was either limited or conflicting.

Since the literature search of Wright et al. (in October 2008) and de Rooij et al. (in July 2015) more research on prognostic factors of hip OA have been conducted, and new methods to assess and review prognostic studies have been developed [5].

Therefore, the aim of this present study was to systematically review the evidence of patient, health, and diagnostic variables associated with the progression of hip OA.

Methods

Search of the literature

A search was made in the databases of Embase, MEDLINE (OvidSP), Web-of-Science, Cochrane Library, PubMed publisher, and Google Scholar from the inception of the database until 14 March 2019, using the keywords hip, osteoarthritis, and prognosis (and their synonyms). We excluded congress abstracts and editorial letters from our search by setting these as limits to restrain the number of found citations without losing valuable citations. The reference lists of relevant articles were screened for additional relevant studies. A complete syntax of the search can be found in Additional file 1. The process of the search was assisted and partly conducted by an experienced medical librarian.

Criteria for selection of studies

The following are the criteria for the selection of studies:

  1. 1)

    The study should investigate the factors associated with the progression of hip OA.

  2. 2)

    The article was written in English, Dutch, German, French, Spanish, Italian, Danish, Norwegian, or Swedish. These languages were sufficiently mastered by at least two reviewers.

  3. 3)

    The article was available in full text.

  4. 4)

    Patients in the study reported complaints like pain, disability, or stiffness of the hip, suspected or confirmed (radiographic or clinical criteria) to originate from OA of the hip.

  5. 5)

    The study design was a cohort or a case-control study or a randomized controlled trial in which the estimation of the prognostic factor was adjusted for the intervention or only investigated in the control group.

  6. 6)

    Progression was determined radiographically or clinically. Radiographic progression could be determined by, for example, X-ray or MRI. Examples of clinical progression were worsening of pain or function or reaching the point of indication for total hip replacement (THR).

  7. 7)

    Follow-up should be at least 1 year (based on the recommendations for measuring structural progression [6]).

  8. 8)

    The study was excluded if the population under investigation had a specific underlying pathology, such as trauma (fractures), infection, rheumatoid arthritis, ankylosing spondylitis, Perthes’ disease, tuberculosis, hemochromatosis, sickle cell disease, Cushing’s syndrome, and femoral head necrosis.

Selection of studies

CHT screened all the titles and abstracts and excluded articles that did not investigate patients with OA of the hip. Secondly, CHT and PAJL independently selected the titles and abstracts using the selection criteria to decide which articles required the retrieval of full text; in case of disagreement, the full text was retrieved. Then, all full texts were independently assessed by CHT and PAJL to include all relevant studies according to the selection criteria. In case of disagreement and both reviewers were unable to reach consensus, SMABZ made the final decision.

Data extraction

Information on the design, setting, study population (e.g., recruitment period, age, gender, definition of hip OA), number of participants, follow-up period, loss to follow-up, prognostic factors, assessment of progression, outcomes, and strength of association were extracted using standardized forms by CHT and checked by PAJL.

Prognostic factors were divided into patient variables, disease characteristics, and chemical or imaging markers. Outcomes were divided into clinical progression, radiographic progression, or (indication for) receiving a THR.

If outcomes were measured at several follow-up moments, all moments were extracted. After the collection of all data, the follow-up moments that were in the closest range to each other were used for the evidence synthesis.

Risk of bias assessment

The quality of all included cohort studies was evaluated using the QUIPS tool [5, 7]. Studies were assessed on six domains: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, and statistical analysis and reporting. An overview of all domains and their items is presented in Additional file 2. Each study was independently scored by CHT and by a second reviewer (DMJD, SMABZ, PKB, JBMRO, or PAJL). In case of disagreement, they attempted to reach consensus; if this failed, a third reviewer (JBMRO or PAJL) made the final decision.

Evidence synthesis

A meta-analysis was considered if clinical heterogeneity was low, with respect to the study population, the risk of bias, and the definition of prognostic factors and defined hip OA progression. In case of a meta-analysis, an adjusted GRADE assessment for prognostic research was used to determine the strength of the evidence [8].

If the level of heterogeneity of the studies was high, we refrained from pooling in the main analysis and performed a qualitative evidence synthesis. Associations were categorized as positive, negative, or no association. Ranking of the levels of evidence was based on Lievense et al. [2] and Davis et al. [9]:

  1. 1)

    Strong evidence: consistent findings (≥ 75% of the studies showing the same direction of the association) in two or more studies with a low risk of bias in all domains of the QUIPS tool

  2. 2)

    Moderate evidence: consistent findings in more than two studies with a moderate or high risk of bias in one or more domains of the QUIPS tool or consistent findings in two studies, of which one study has a low risk of bias in all domains of the QUIPS tool

  3. 3)

    Limited evidence: one study with a low risk of bias in all domains of the QUIPS tool or two studies with a moderate or high risk of bias in one or more domains of the QUIPS tool

  4. 4)

    Conflicting evidence: < 75% of the studies showing the same direction of the association

If a prognostic factor was described in two different articles that investigated the same study cohort and outcome of progression, one study was selected to include in the evidence synthesis. In this case, we selected the article according to a decision tree: (1) lowest risk of bias, (2) prognostic factor is the primary outcome of the study, and (3) the largest number of participants.

Post hoc changes to the study protocol

After contact with one of the developers of the QUIPS tool, we learned that it is not validated to judge the risk of bias of case-control studies and would probably not adequately take into account the higher risk of recall bias and the selection bias of case-control studies. Therefore, we decided to exclude case-control studies from our evidence synthesis, except for nested case-control studies. Nested case-control studies are less prone to selection and recall bias because of the underlying known cohort [10], which can be judged using the QUIPS tool.

Results

Included studies

The initial search yielded 6429 citations of which 57 articles were finally included. Figure 1 shows the reasons for the study exclusion, and Table 1 presents a brief overview of the characteristics of the 57 included studies (a more extensive overview is available in Additional file 3). Of the 57 studies, 48 were cohort studies (37 with a prospective design), 4 were nested case-control studies, and 5 were case-control studies. These last 5 studies were excluded from the evidence synthesis for the reasons mentioned above.

Fig. 1
figure 1

Flowchart of the search and selection of studies

Full size image
Table 1 Characteristics of the selected studies
Full size table

Risk of bias assessment

In 68% of all assessed domains from all studies, there was an immediate consensus between the reviewers (Cohen’s kappa 0.375, linear weighted kappa 0.484). In 9 assessments of a domain (3%) in 6 different studies, a third reviewer made the final judgment. In total, 15 studies scored a low risk of bias in all domains [15, 16, 21, 29, 30, 32, 34, 37, 41, 44, 47, 49, 53, 57, 63] (Table 2).

Table 2 Risk of bias assessment summary (QUIPS)
Full size table

Prognostic factors

We identified 154 possible prognostic factors: 23 patient variables, 77 disease characteristics, and 54 chemical markers or imaging markers. Fifty-one factors were only investigated once in a single cohort or study (not a low risk of bias study) and could not be included in the evidence synthesis. An overview of all the results and risk of bias assessment of the studies describing these factors is presented in Additional file 4. The remaining 103 factors were included in the evidence synthesis. To decrease heterogeneity, evidence synthesis was done separately per group of outcomes (radiological progression, clinical progression, or THR). However, heterogeneity was still considered high in each outcome group, mainly within respect to the definition of the prognostic factor, progression, and measure of the association. Therefore, we refrained from pooling and performed a best-evidence synthesis. If a factor could not be subdivided because it was described by two or three studies that used a definition of progression, all in a separate group of outcome, we combined the groups of outcomes. The results of these factors are presented in Additional file 5.

Evidence for factors predicting progression

Strong evidence was found for a higher K-L grade at baseline, superior or (supero) lateral femoral head migration, and subchondral sclerosis to be predictive of faster progression to THR or more patients progressing to THR. Body mass index was found not to be predictive of faster or more progression to THR (Table 3).

Table 3 Factors predicting (indication for) total hip replacement (THR)
Full size table

Strong evidence was found for no association between radiological progression and the following markers: C-terminal telopeptide of collagen type I (CTX-I), cartilage oligomeric matrix protein (COMP), N-terminal telopeptide of collagen type I (NTX-I), and N-terminal propeptide of procollagen type I and type III (PINP, PIIINP) (Table 4).

Table 4 Factors predicting radiological progression
Full size table

Strong evidence showed comorbidity to be predictive of clinical progression. On the other hand, gender, social support, use of pain medication at baseline, quality of life at baseline, and limited range of motion of internal hip rotation or external hip rotation were not predictive of clinical progression (Table 5).

Table 5 Factors predicting clinical progression
Full size table

For other factors, only moderate, limited, or conflicting evidence was found for predicting or not predicting progression (Tables 3, 4, and 5).

Discussion

In this study, we systematically reviewed all 154 factors predictive of progression of hip OA, reported in 57 studies. Compared to earlier reviews, there was a considerable amount of additional evidence available for the factors previously reported in reviews, as well as evidence for factors not earlier described.

In this review, some results had changed compared to the review of Lievense et al. in 2002 [2]. Firstly, because of the new evidence emerging from the later studies, especially studies with a clinical outcome of progression. Secondly, because we used a different method to assess the risk of bias, some studies were no longer considered to have a low risk of bias. The QUIPS tool seems to apply stricter criteria than the method used by Lievense et al. in 2002. Thirdly, we divided the outcomes into three different groups of progression. Thus, due to these methodological differences (together with additional studies), we were unable to confirm an atrophic bone response as a predictor for radiological progression or progression to THR. On the other, we were able to confirm their conclusion on BMI as not predictive of progression and faster progression in patients with a superolateral migration of the femoral head.

Most of the prognostic factors reported by Wright et al. in 2009 [3] were confirmed in this present review in one or more of the outcome groups. The differences found in age, femoral and acetabular osteophytes, and hip pain at baseline were (as with Lievense et al.) a combination of new evidence, differences in the risk of bias assessment, and the division into defined groups of progression. The study from de Rooij et al. in 2016 [4] reviewed the evidence for predictors of the course of pain and function and found comorbidity and vitality (SF-36) to be predictive of function, as we found for clinical progression. However, although they also used the QUIPS tool to assess the risk of bias, they used a different cutoff point to classify a study as having a low risk of bias. Therefore, some earlier findings of strong evidence for no association with the course of pain or function were confirmed as only moderate evidence for no association with clinical progression in our review. Other differences between this review and the present one are mainly attributable to the differences in the selection criteria. In Table 6, we summarized all factors with strong evidence to be predictive of progression found in one of these four reviews and the overlap and differences in evidence for these factors.

Table 6 Overview of factors with strong evidence to be predictive of progression, overlap and differences between this review and the review of de Rooij et al., Wright et al., and Lievense et al.
Full size table

Strengths of this present review are the sensitive literature search and our systematic approach to the selection, risk of bias assessment, and the best-evidence synthesis. Therefore, we have presented an extensive overview of reported prognostic factors and existing evidence for their associations. In performing the evidence synthesis divided into outcome (radiological, clinical, or THR), we decreased the heterogeneity and we believe the results to be more accurate for daily practice. Unfortunately, heterogeneity was still too high to perform a meta-analysis. Therefore, we were bound to a best-evidence synthesis and unable to calculate the strengths of the associations. This limits the translation to the daily clinical practice. Another disadvantage of this synthesis compared to a meta-analysis is that smaller studies contribute to the result with the same weight as larger studies, even though the smaller studies may have low power to show a statistically significant association.

In the selection of studies, several restrictions were imposed. First, languages were restricted to ensure that at least two researchers had a reasonable understanding of the languages included so all articles were reliably assessed. However, this implies that we may have missed studies from countries in which publication in English is less common. Secondly, negative results (i.e., no association was found) are less likely to be published and are therefore not well represented in this review.

We used the QUIPS tool to assess the risk of bias. Nine other studies using this tool reported an inter-rater agreement ranging from 70 to 89.5% (median 83.5%) and a kappa statistic ranging from 0.56 to 0.82 (median 0.75) [7]. Compared to these data, our inter-rater agreement was low and considered to be moderate. Disagreement was mainly due to the differences in interpretation of items of the QUIPS tool; however, only for very few items, a third reviewer was needed to make a final decision.

Hip dysplasia and femoral acetabular impingement were initially considered to be underlying pathologies and were excluded from this analysis. However, the range of severity of these morphologies is substantial, i.e., some of these morphologies should clearly be considered as an underlying pathology, whereas others are more subtle and sometimes undiagnosed. These subtle morphologies might be considered to be possible prognostic factors, rather than underlying pathologies. Therefore, all citations were screened using the terms “hip dysplasia” and “femoral acetabular impingement” in the title or abstract. However, we found only one small study [35] which investigated the radiographic findings of femoral acetabular impingement as a prognostic factor (results of this study are included in Additional file 4). In the studies already included, three studies did not specifically include patients with hip dysplasia or femoral acetabular impingement but did investigate the associated angles (Wiberg’s center edge angle and alpha angle, respectively). Since the evidence for these associations with the progression of hip OA was weak, future studies and reviews should investigate these morphologies as possible prognostic factors.

Conclusion

We conclude that there is consistent evidence that four factors (comorbidity, K-L grade, superior or (supero) lateral femoral head migration, and subchondral sclerosis) were predictive of progression of hip OA, whereas 12 factors were not predictive. The evidence for other factors was weak or conflicting. Health professionals caring for patients with hip OA will benefit from the insight in prognostic factors, e.g., patients more likely to progress rapidly may need an intensified symptomatic treatment or early referral to an orthopedic surgeon. For this, we still need more high-quality research focusing on the prognostic factors in hip OA.