The effect of variation in interpretation of the La Trobe radiographic foot atlas on the prevalence of foot osteoarthritis in older women: the Chingford general population cohort

Osteoarthritis (OA) is an important cause of global disability, with adult prevalence rates reported between 8.5–22.0% for symptomatic radiographic knee OA [1‐3], 3.4–8.9% for symptomatic radiographic hip OA [2, 4, 5]. The prevalence of radiographic hand OA has been reported to range from 27.0 to 83.0% [2, 6]. Hand OA is said to consist of several phenotypes that make it more complex to study [7]. Whilst investigation of foot joints may be more aligned to those of the hand as a peripheral joint site with multiple small bones and joints, the prevalence of radiographic foot OA is much less understood.

Foot pain is often linked to foot OA and is highly prevalent in the general population, with estimates that range between 15.0–63.0% [8‐11]. Although conventional radiographs have been used traditionally to assess OA there is discordance in how radiographic and symptomatic OA are defined [12‐14] and a lack of methodological standardisation across studies [9]. For investigations of foot OA, issues such as the considerable variation in study populations, the radiographic views taken, which foot joints are examined, the grading systems applied and definitions for prevalence of radiographic foot OA are highlighted as potential reasons for the lack of conclusive data regarding radiographic and symptomatic foot OA [15]. Of these factors the lack of standardisation in the methods used to assess radiographic foot OA [15], the number of foot joints included to define foot OA [16] and the disparity between radiographic OA and symptomatic OA [17, 18] appear to be key issues to address. Recently, the UK population prevalence of symptomatic radiographic foot OA has been estimated as 16.7% in adults aged over 50 years [19] and in the US prevalence estimates of pain at specific foot locations range between 7 and 13% in adults (30–100 years) [20].

Experts agree that the separate grading of osteophytes (OPs) and joint space narrowing (JSN) using standardised and validated atlases is an important way forward [21, 22]. In an attempt to address this, Menz et al. [23] developed a radiographic atlas specifically for standardising the documentation and interpretation of foot OA. The atlas uses an ordinal scale to score the presence of OP and JSN at five joints within the foot on dorsoplantar and lateral views [23]. Previously investigators largely relied on the Kellgren and Lawrence classification system [24] to define OA in individual foot joints, which was often limited just to the first metatarsophalangeal joint (1stMTPJ) [12, 15, 25].

Menz et al. reported good intra-rater reliability (percentage agreement from 86.0 to 99.0% and weighted κ from 0.45 to 0.95), of the La Trobe Foot Atlas (LFA) and construct validity relative to foot symptoms [16, 23]. The LFA has since been used to determine the prevalence of radiographic OA at the global foot level in relation to foot pain [19] and effects of intervention at an individual joint level [26]. Studies using the LFA that did not include a member of the original team that developed the atlas are scarce [25] or do not discuss the use of the atlas [27] such that the interpretation of the LFA scoring has yet to be evaluated.

The presentation of radiographic features varies quite widely. As radiographic atlases use semi-quantitative or ordinal grading systems to classify individuals, often into 4 or 5 categories, a degree of interpretation is required in order to categorise OA features [28‐31]. We postulated that, as with other radiographic atlases, the LFA ordinal technique for scoring introduces an interpretative approach, that may potentially lead to an over or under-estimation in the prevalence of OA [31]. This is particularly likely when an unclear view of a joint is being assessed, which happens often for views of the midfoot and certain hind-foot joints [16, 23, 32]. The authors of the original LFA themselves do suggest from their inter-rater reliability results that “there is some degree of inherent variability in the interpretation of some aspects of the atlas” [23]. We wished to evaluate how much this variation in interpretation can affect the prevalence of radiographic foot OA.

When the foot radiographs (Chingford year 23) were scored using the LFA (Technique 1), the total (i.e. combined joints of left and right feet) prevalence of radiographic foot OA in any joint in the right and left foot was 81.2% using only the dorsoplantar view and 83.5% using only the lateral view. When scores were combined for both views and both feet radiographic foot OA was present in 89.9% of participants (Table 2). For Technique 2 (categorising joints that were difficult to grade as ‘missing’) the prevalence of radiographic foot OA was 83.5% (both feet, both views). For Technique 3 (attributing an over estimated score to joints that were difficult to grade) the prevalence of radiographic foot OA was 97.2% (Table 3).

At the individual joint level, Technique 2 elicited a lower presence of radiographic foot OA than Technique 1 (Table 2). With the exception of the 2^nd CMJ (both feet and both views) that elicited a difference of 36.2% (both feet) between Techniques 1 and 2 joint scores, all other joint scores were within an acceptable range (left foot: 1.4–4.6%; right foot: 0.4–6.3%). Conversely, at the individual joint level Technique 3 elicited a higher presence of radiographic foot OA than Technique 1. With the exception of the N1stCJ (both feet, dorsoplantar view) that elicited a difference of 49.1% (left foot) and 52.3% (right foot) between Technique 1 and 3 scores, all other joint scores were within a less wide range (left foot: 5.1–19.3%; right foot: 4.2–21.6%).

At the individual joint level, using Technique 1, the presence of radiographic foot OA for combined OP and JSN was higher with a wider range (18.3–74.3%) than Technique 2 (17.9–46.3%). At the individual joint level, using Technique 1, the presence of radiographic foot OA for combined OP and JSN was lower with a wider range (18.3–74.3%) than Technique 3 (39.9–79.4%).

In this study, we sought to extend knowledge of radiographic foot OA by examining three different interpretive approaches to classifying foot OA using the LFA. The three different ways of interpreting the LFA scoring system when scoring individual joints that we used is technically difficult and each resulted in different estimates of foot OA prevalence at both the individual joint and global foot level.

Similar to other radiographic scoring methods, such as Kellgren and Lawrence [24, 39] and the Osteoarthritis Research Society International (OARSI) atlas [40], there is potential ambiguity in the interpretation of the scoring for OPs and JSN within individual joints using the LFA. Scoring of foot joints on radiographs presents specific problems due to overlap of bones that makes it difficult to clearly see the joint line and OP on any one view in all joints of interest. Through comparison of the different techniques we showed the potential for the range of prevalence estimates of person level radiographic foot OA to be between 83.5% and 97.2%.

Menz et al. [16] reported the prevalence of radiographic foot OA in their elderly sample (as 93%, which is similar to our standard LFA assessment of 89.9% and within our range when utilising the two additional techniques. Menz et al. [16] also reported a joint-specific prevalence rate for individual joints that ranged between 23.0–60.0% which is similar to the range between 18.3–74.3% that we found. The sample size that Menz et al. [16] investigated and age was similar to ours (n = 197, mean age 75.9 years, [SD] 6.6), however they were drawn from a retirement village and a university health sciences clinic in Melbourne, Victoria, Australia with 64.0% women, whilst ours were all women drawn from a general population in the UK.

Other investigators have reported lower prevalence estimates for foot OA. For example, in an American population, the Clearwater Osteoarthritis Study, a prospective cohort consisting of 3463 participants (40–94 years), Wilder et al. [41] reported a prevalence of 20.0% of radiographic foot OA. Within that study, the focus was on one only foot joint only, the 1st MTPJ, so a lower prevalence of OA at the individual foot joint level is expected. Our findings were higher 35.8% (left) and 42.2% (right) for presence of OA in the 1st MTPJ. The lower estimate produced by Wilder et al. [41] may be due to the fact that their scoring was based on the traditional Kellgren and Lawrence scale which is not as sensitive to radiographic foot OA as the LFA [16].

It is not just the approach that is open to interpretation. Even using the different techniques, our estimates are much higher, than the most recent UK study that estimated the population prevalence of symptomatic radiographic foot OA as 16.7% [19]. The latter study used foot pain and foot OA (ie symptoms plus radiographs) to define their prevalence of symptomatic foot OA, whereas we only used foot OA (radiographs). This highlights the marked difference in prevalence estimates dependent on whether the focus of investigation is on symptomatic radiographic foot OA or just radiographic foot OA, the latter being distinctly much higher [2]. The difference in prevalence estimates due to the case definition has been noted in OA at other joints sites [2].

Each of these examples may go some way to explaining the variation in published prevalence estimates of radiographic foot OA, especially when different techniques are employed and different joints included. Other factors that may explain the differences in prevalence estimates of radiographic foot OA could be related to the subjectivity of the scoring method being ordinal as opposed to objective measurements such as joint space width. As a further example, we found that scoring may be confounded as the individual features of OP or JSN are not presented separately but are mixed and this may distract the scorer to judge the “best-fit” picture due to the overall appearance rather than to just the OP or JSN they are scoring.

The advice given in the LFA indicates that use of both dorsoplantar and lateral views is ‘gold standard’ and should be applied where possible to ensure an appropriate level of sensitivity to OA [23]. Further evaluation of the LFA has shown that good sensitivity (94.6%) can be obtained in the 1st MTPJ when only a dorsoplantar view is available. However, substantially lower sensitivity was achieved for the other joints (between 31.0 and 60.7% of cases) [16]. The 1^st MTPJ is the largest of the MTPJs and is not obscured by other joints when observed in radiographs and as such easier to assess the presence of OPs and JSN. Menz et al. [16] reported the combined view was 42.4% for the 1^stMTPJ which is very similar to our estimate of combined view 1^st MTPJ presence of OA as 35.8% for the left foot (dorsoplantar view: 27.1%; lateral view: 22.9%) and 42.2% for the right foot (dorsoplantar view: 33.0%; lateral view: 27.5%). Of note, the joints that showed most difference between our techniques were the 2nd CMJ and N1stCJ. These joints are also the ones noted to be difficult to score in the LFA atlas due to considerable amount of overlap of bones and joints [23].

There are limitations to this study. Firstly, it is possible that our estimates of prevalence may have been confounded by the lower reproducibility of the rater in this study than that of the original authors of the LFA [23]. There are a number of explanations for lower reliability scores in our preliminary work such as the foot positioning for the reliability study differed from that undertaken by the LFA, availability of only one view (non-weight-bearing dorso-plantar) and lower quality of foot radiographs versus higher quality of resolution of electronic images used in year 23. For the development and testing of the LFA, the same authors selected the radiographs for each LFA classification grade on which their reliability was calculated [23]. This may provide more stable predictions of reliability scores but may not be as readily applicable to new raters external to the original development team.

Secondly, the cohort used in the development of the LFA was a sample of the Australian population over the age of 65 years, whereas the foot radiographs used in this study were all from a sample of women of the UK population aged 69–93 years. There is currently no available foot radiographic data that compares different populations that may have different physiology, anatomy and genetics. Consequently, we do not know how representative the pictures used to explain the scoring method within the LFA are for global comparisons or how closely the Chingford 1000 Women’s study cohort foot radiographs may align to them. Of note, within the LFA there are not separate pictures for OPs or JSN for men and women. It is currently not known if factors such as joint width are smaller in foot joints of women than men which may affect interpretation and scoring for OA.

Thirdly, differences in prevalence estimates of foot OA could be related to study populations and the focus of the investigation. The focus of our investigation was radiographic features of foot OA only. We have not aligned this to symptoms of foot pain as our aim was to evaluate the scoring technique for foot OA using a validated radiographic atlas. Our findings are therefore not directly comparable with other investigators reporting on the prevalence of symptomatic foot OA. Whilst symptomatic foot OA may be more prevalent in women [19] we are aware that the prevalence of foot OA was very high in our study. We believe this could be due to a combination of the population being all women aged over 69 years in whom OA has generally been found to be more prevalent [42]. Additionally, in our study, OA was defined radiographically which has been shown to lead to higher estimates than other definitions such as ‘self-reported OA’ and ‘symptomatic OA’ (combined radiographic OA with symptoms) [2]. Estimates of the prevalence of OA of a similar order have been reported at other peripheral joints sites in other populations of older women [43].

This study supports the use of the La Trobe Foot Atlas to facilitate standardised scoring of foot OA in existing current and historical radiographs of established large population cohorts with the caveat that the interpretative scoring technique requires acknowledgment. We have evaluated three different ways of interpreting the scoring system when scoring of individual joints is technically difficult and results in different estimates of foot OA prevalence at both the individual joint and global foot level. This strengthens the case for further refinement of definitions for foot OA between investigators and improved comparability between studies. Future work should focus on agreement on the best strategy to improve comparability between studies to begin to identify the risk factors for foot OA. From that the field can move forward in developing best clinical strategies for prevention and management of foot OA.