Medical imaging for plantar heel pain: a systematic review and meta-analysis

The protocol of this systematic review was prospectively registered with PROSPERO (Registration No. CRD42020172398) and has been reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [15].

Searches were conducted in MEDLINE, CINAHL, SPORTDiscus, Embase and the Cochrane Library from inception to 12th February 2021 – Additional file 1. Citation tracking using Google Scholar was performed to identify any further relevant citations. Reference lists were screened for studies not identified in the initial search.

Eligible articles were peer-reviewed studies published in the English language. Studies had to be cross-sectional observational studies that compared medical imaging findings from a group of adult participants with PHP to an independent control group of adult participants without PHP.

Studies were excluded if they exclusively compared a symptomatic foot with the contralateral asymptomatic foot of the same participant (e.g. no independent control group comparison) – this was done to avoid confounding where the condition may have been developing in the contralateral foot but was still asymptomatic. Studies were also excluded if they included participants who had any self-reported inflammatory arthritis (e.g. seronegative arthropathy), endocrine/neurological condition (e.g. diabetic peripheral neuropathy), surgery (e.g. joint fusion), or trauma (e.g. major fractures) that had affected lower limb sensation or their ability to walk/run and if relevant to the imaging modality of interest. The same exclusion criteria were applied to the control group without PHP in each study, who were also required to be asymptomatic of PHP on both feet.

The search results were exported from the bibliographic databases into Endnote X9 (Thomson Reuters, New York, USA) and duplicate citations were removed. Two authors (CD and JC) examined all the study titles and abstracts independently, and studies deemed ineligible were excluded. The full text articles of the remaining studies were obtained and examined against the eligibility criteria for inclusion in the systematic review. If consensus agreement could not be agreed between the two authors, a third author (KL) was consulted to resolve the disagreement.

A data extraction form was implemented to extract the individual study characteristics (e.g. BMI) and the imaging modality (e.g. ultrasound). The primary variables of interest included: plantar fascia thickness on ultrasound and MRI, hypoechogenicity on ultrasound, plantar fascia tear on ultrasound and MRI, plantar fascia stiffness on sonoelastography, hyperintensity on MRI, hyperaemia on power Doppler ultrasound, plantar intrinsic muscle size on ultrasound and MRI, plantar calcaneal spur on x-ray, bone marrow oedema on MRI, calcaneal crescent sign on MRI, and radioisotope uptake on scintigraphy. Variables that could have led to bias were also extracted (e.g. blinding). Two authors (CD and GW) independently extracted and compared their data to minimise errors. A third author (KL) was consulted when consensus on the data extracted could not be reached.

The National Institutes of Health (NIH) quality assessment tool for observational cohort and cross-sectional studies was used to assess study quality and risk of bias [16]. The tool has 14 questions that are specific to cross-sectional studies (the study type included in our review), which encompass the key concepts required to investigate the internal validity of a study (selection, information, measurement and confounding bias). The tool allows a rating to be applied to a study (rated as ‘poor’, ‘fair’ or ‘good’) based on individual details and consideration of the concepts, rather than a tally scoring system. Low risk of bias equates to a ‘good’ quality rating, whereas high risk of bias equates to a ‘poor’ quality rating. Two authors (CD and MK) independently performed the quality assessment and disagreements were resolved through consensus. A third author (KL) was to be consulted when consensus could not be reached, however this was not required.

Meta-analyses were performed using Review Manager (RevMan, Version 5.4, The Cochrane Collaboration, 2020). Due to the variation in study methods, all meta-analyses were conducted using an inverse-variance random-effects model. Statistical heterogeneity between studies was examined using I² and Chi² statistics. The I² statistic describes the variability in effect estimates that may be apportioned to study heterogeneity and is displayed as a percentage value where 0% to < 30% might not be important; 30% to < 60% may represent moderate heterogeneity; 60 to 90% may represent substantial heterogeneity, and > 90% may represent considerable heterogeneity [17]. Chi² statistics were deemed statistically significant for heterogeneity when p < 0.1, although it is recommended to base analysis models on a thorough examination of heterogeneity rather than solely on one statistic [17].

Continuous outcome variable data were analysed by inputting each individual study’s mean outcome values, standard deviation (SD) and sample size for the PHP and control groups. The mean difference between groups and 95% confidence interval (CI) were calculated and a weighted pooled estimate for the individual studies was obtained. Dichotomous outcome variable data were analysed by inputting each individual study’s number of events and sample size for the PHP and control groups. The odds ratio (OR) and 95% CI were calculated, and the inverse variance method applied in order to determine the weighted pooled estimate. Where a study reported no events in both groups, a continuity correction was used based on a function of the reciprocal of the opposite group [18].

For meta-analyses that included both studies that used blinded assessors and studies that used unblinded assessors (i.e. assessors were aware whether or not participants had PHP), sensitivity analyses were performed to assess for potential assessor bias where appropriate. Where studies reported unilateral foot data and where the symptomatic foot of the PHP participants could be compared to the same side in the control group (i.e. left vs left or right vs right), then the most conservative data were used for the purpose of meta-analysis.

The database search identified a total of 2973 unique citations of which 42 studies met the eligibility criteria for inclusion in the review [19‐60] – Fig. 1. The excluded studies and the reasons for exclusion following full text article assessment are presented in Additional File 2. There was a total sample size of 2928 participants; 1367 PHP participants (62% female, mean age 46 years) and 1561 control participants (56% female, mean age 42 years).

Overall, 16 of the 42 (38%) studies reported if assessors were blinded to whether participants had or did not have PHP, BMI was not recorded in 21 (50%) studies, bilateral heel data (where participants’ had PHP on both feet and/or where both asymptomatic feet of the controls were included) was included in 25 (60%) studies, and the population from which the sample was recruited was not recorded in 23 (56%) studies – Table 1.

All 42 studies were appraised using the NIH quality assessment tool, with 16 (38%) rated poor, 17 (41%) rated fair, and 9 (21%) rated good (Table 2). Details of the quality appraisal for each study are included in Additional file 3.

Measurements of plantar fascia thickness were reported in 31 studies, and of these, 26 used ultrasound alone [19‐44], 4 used MRI alone [46‐49], and 1 used ultrasound and MRI [45].

Three unblinded ultrasound studies reported the number of participants with plantar fascia thickness > 4 mm [19, 22, 35]. Meta-analysis was conducted and included a total of 99 PHP participants and 160 control participants. PHP participants were greater than 600 times more likely to have a plantar fascia thickness > 4 mm compared with control participants (OR 634.12, 95% CI 38.57 to 10,424.05, p < 0.001) – see Fig. 5. Heterogeneity was found to be moderate for this analysis (Tau² = 3.38; Chi² = 4.48, I² = 55%). One other ultrasound study reported a different cut-off for plantar fascia thickness of > 4.5 mm [38], and therefore was not included in the meta-analysis. It found 73 (91%) of the PHP participants had plantar fascia thickness > 4.5 mm compared to 4 (2%) of the control participants.

One MRI study reported the number of people with plantar fascia thickness > 4 mm [49]; this study found 15 (75%) of the PHP participants and none (0%) of the control participants presented with this finding.

Three studies measured hyperaemia using power Doppler ultrasonography [23, 35, 41]. A meta-analysis could not be conducted due to heterogeneity between studies (e.g. variation in study design and outcome measures). One study purposively sampled a PHP group with fascia thickness > 5 mm [41]. Two studies graded hyperaemia using a 1–4 scale [35, 41], and a comparison of the number of people with hyperaemia between PHP and control participants in these two studies is presented in Table 3. The third study [23], measured vascular index using power Doppler ultrasonography with increased vascularity in the PHP group (mean 2.4, SD 1.4) compared to the control group (mean 1.6, SD 0.4).

Seven studies measured elasticity of the plantar fascia using sonoelastography [24, 28, 37, 38, 44, 50, 51]. Two studies excluded symptomatic participants with abnormal features on standard ultrasound [50, 51]. A meta-analysis could not be conducted due to heterogeneity between the studies (e.g. study design and sonoelastographic variables measured). These studies generally found that the plantar fascia was softer or less stiff. A summary of individual study results is presented in Table 4.

Six studies recorded the presence of plantar fascia tears [19, 20, 22, 34, 45, 49]. Four studies used ultrasound alone [19, 20, 22, 34], one study used MRI alone [49], and one used both ultrasound and MRI [45].

Five studies measured heel fat pad thickness [21, 34, 39, 45, 53], one of which reported measurements using both ultrasound and MRI [45].

Two studies measured muscle size; one study measured cross-sectional area and muscle thickness of the abductor hallucis muscle using ultrasound [31], and one study measured intrinsic foot muscle volume, which was normalised to body mass using MRI [54].

The ultrasound study found no significant difference (p = 0.45 – the authors presented p-values to 2 decimal places only) in abductor hallucis muscle cross-sectional area between the PHP participants (mean 2.00 cm², SD 0.52) and the control participants (mean 1.87 cm², SD 0.47). There was also no significant difference (p = 0.46) in abductor hallucis muscle thickness between the PHP participants (mean 1.16 cm, SD 0.23) and the control participants (mean 1.10 cm, SD 0.24) [31].

The MRI study measured muscle volume in three areas; total intrinsic foot muscle volume, rearfoot muscle volume, and forefoot muscle volume (all of which were normalised to body mass) [54]. Firstly, PHP participants had 245.3 mm³/kg less total intrinsic foot muscle volume compared to control participants. PHP participants had a mean of 1838.0 mm³/kg (SD 277.1) and control participants had a mean of 2083.3 mm³/kg (SD 258.7). This difference was not statistically significant (it approached significance p = 0.056), but the Cohen’s d effect size was large at 0.92. Secondly, PHP participants had 195.5 mm³/kg less rearfoot volume compared to control participants. PHP participants had a mean of 746.8 mm³/kg (SD 129.18) and control participants had a mean volume of 942.5 mm³/kg (SD 208.02). This difference was statistically significant (p = 0.023) and the Cohen’s d effect size was large at 1.13. Thirdly, PHP participants had 49.6 mm³/kg less forefoot volume than control participants. PHP participants had a mean volume of 1091.2 mm³/kg (SD 169.51) and the control participants had a of 1140.8 mm³/kg (SD 149.48). This difference was not statistically significant (p = 0.496) and the Cohen’s d effect size was moderate at 0.31.

Six studies measured the presence of calcaneal spur using x-ray [20, 55‐59]. Only two of the studies used assessors that were blinded [56, 59]. Meta-analysis of the six studies was conducted with a total of 326 PHP participants and 846 control participants. PHP participants were almost 5 times more likely to present with a calcaneal spur compared to control participants (OR 4.92, 95% CI 2.12 to 11.39, p < 0.001) – see Fig. 12. Heterogeneity was found to be substantial for this analysis (Tau² = 0.84; Chi² = 26.85, I² = 81%).

A sensitivity analysis of the two studies that used blinded assessors was conducted (PHP n = 129, control n = 319). Compared with the overall analysis (presented above), the sensitivity analysis of studies that used blinded assessors found greater odds of having a calcaneal spur in the PHP participants (OR 12.19, 95% CI 5.01 to 29.65, p < 0.001) – see Fig. 12. This finding was also higher than the analysis of studies that used unblinded assessors (OR 3.13, 95% CI 1.03 to 9.53, p = 0.040). Heterogeneity was found to be moderate for the blinded analysis (Tau² = 0.19; Chi² = 1.78, I² = 44%) and substantial for the unblinded analysis (Tau² = 1.00; Chi² = 18.17, I² = 83%).

Two studies measured the presence of bone marrow oedema within the calcaneus using MRI [46, 49], with one study using assessors that were blinded [49]. A meta-analysis was not conducted as there were no events in either the PHP or control group in one study [46]. The other study found that 7/20 PHP participants (35%) presented with bone marrow oedema in the calcaneus compared to 0/20 (0%) in control participants [49].

One study measured the cross-sectional area and volume of the calcaneal tuberosity trabeculae (calcaneal crescent) using MRI [47]. This study found that PHP participants had greater cross-sectional area of the calcaneal crescent when compared with control participants (mean 100.2, SD 29.2 versus mean 73.7, SD 18.9 mm², p = 0.019), greater volume, (mean 3.06, SD 1.10 versus mean 1.99, SD 0.68 cm³, p = 0.006), and lower contrast-to-noise ratio (mean − 38.1, SD 11.0 versus mean − 28.4, SD 13.0, p = 0.009).

Two studies measured radioisotope uptake in the calcaneus using technetium-99 m bone scans [55, 60]. A meta-analysis was not conducted as one study did not report the phase of the bone scan that observations were recorded [55], and the other study did not report the number of control participants [60]. One study reported increased uptake in the calcaneus in 16/22 (73%) in PHP participants and 0/17 (0%) in control participants [55]. The other study reported increased uptake in 31/52 (60%) of PHP participants and 0% (number not reported) in control participants [60]. In both studies, no statistical comparisons of the radioisotope uptake differences between the groups were made.

The imaging features outlined above are consistent with degenerative soft tissue changes characterised by fascia thickening, fascia tissue changes, presence of fascia tears, and loss of fascia elasticity. A thickened plantar fascia with degenerative changes is consistent with our previous systematic review [14].

Regarding plantar fascia thickness, meta-analysis of ultrasound studies found that participants with PHP had a mean proximal plantar fascia thickness that was 2.00 mm thicker than control participants. However, a sensitivity analysis found a lesser but still thicker difference of 1.62 mm for blinded studies compared to 2.20 mm for the unblinded studies. This suggests that unblinded studies, with a higher risk of assessor bias, may have over-estimated plantar fascia thickness in participants with PHP. Therefore, we have elected to focus on the more conservative interpretation that people with PHP have a plantar fascia that is 1.62 mm thicker on average than people without PHP (we have done this for all other findings in the discussion below). Meta-analysis of MRI studies found that participants with PHP had an even thicker plantar fascia (3.17 mm thicker) than control participants; although, there were only 2 studies in the MRI analysis compared with 21 studies in the ultrasound analysis. MRI thickness measurements can over-estimate tissue thickness measurements as it is dependent on the orientation of the slice from which the measurement is taken, and if that slice is oblique to the plane of maximum thickness, it can measure the tissue to be thicker than it actually is. This has been found in other populations and elsewhere in the body [61]. Accordingly, ultrasound measurements of tissues like the plantar fascia are generally more accurate.

Plantar fascia thickness changes can also be categorised by dichotomising participants into those with a plantar fascia that is thicker than 4 mm and those with a plantar fascia that is 4 mm or less [10]. Meta-analysis revealed that participants with PHP were 634 times more likely to have a plantar fascia thickness greater than 4 mm when compared with healthy controls. However, this finding should be interpreted with the knowledge that all studies in the analysis used assessors that were not blinded, and that two [19, 22] of the three studies used paired heel data from the same participants if they had bilateral PHP. Paired data can be used to increase sample size, however it can lead to reduced variability in the sample and result in statistically significant findings that may be spurious [62]. Nevertheless, it can be concluded that the plantar fascia is thicker in people with PHP on both ultrasound and MRI, and the odds of the fascia being thicker than 4 mm is greatly increased on ultrasound.

Not only does the fascia thickness increase in people with PHP, but tissue changes within the fascia can also be detected with medical imaging. The presence of plantar fascia hypoechogenicity on ultrasound and hyperintensity of the signal on MRI were found to be significantly associated with PHP. Participants with PHP were nearly 31 times more likely to have hypoechogenicity on ultrasound and 146 times more likely to have signal hyperintensity on MRI of the plantar fascia. Accordingly, people with PHP are substantially more likely to show signs that are consistent with degeneration of the plantar fascia on ultrasound and MRI as detected by hypoechogenicity and hyperintensity, respectively.

In addition to the plantar fascia tissue changes outlined above, we were interested in whether plantar fascia tears were more likely in people with PHP, which was not analysed in our previous review [14]. Meta-analysis found no significant differences between people with and without PHP for the presence of a plantar fascia tear on both ultrasound and MRI. However, both meta-analyses had relatively low sample sizes, and as a consequence, the OR estimates had wide confidence intervals, so more studies are needed for this analysis to improve the precision of the estimates, which is needed to know definitively if plantar fascia tears are truly associated with PHP. In addition, studies included in this analysis provided unclear definitions of a tear on imaging, and some may have assessed for a full tear only, as opposed to both partial and full tears. A tear within the fascia, whether partial or full, is of clinical interest, as it may correspond to an acute episode where the patient remembers an incident that triggered the pain and is worth considering during treatment as greater weightbearing relief may be necessary for healing to occur. We believe this imaging feature needs further investigation with a strict definition of what constitutes a tear.

While there is clear evidence for changes in the plantar fascia tissue in people with PHP on ultrasound or MRI, such as thickness or structural changes, findings from some other modalities are less convincing at this stage. Sonoelastography studies included in this review suggest a loss of elasticity in the fascia in those with PHP. Two of the studies reported this feature occurring in isolation without other plantar fascia changes [50, 51], which suggests there might be the potential for early diagnostic ability with sonoelastography, however it is currently unknown whether such a finding is clinically worthwhile from a management perspective. A meta-analysis could not be conducted due to differences between studies in the sonoelastographic variables measured, therefore findings from sonoelastography studies could not be synthesised or summarised. Despite sonoelastography being of interest in PHP research, future studies of PHP using sonoelastography need improvement; that is, methods and measurements need to be standardised.

There may also be differences in plantar intrinsic muscle size between PHP and healthy controls, but again, the lack of studies prohibited a meta-analysis of this. Indeed, in two studies, intrinsic foot muscle size (cross-sectional area and volume) was found to be decreased in participants with PHP. Our findings are essentially the same as those of Osborne and colleagues [63] who conducted a systematic review that was specific to muscle strength and size in people with and without PHP; that is, they did not investigate wider medical imaging findings. One issue when considering muscle size from cross-sectional studies is that causality cannot be inferred, so even if people with PHP have smaller intrinsic muscles, for example, it cannot be determined if the decrease in size is the cause of PHP or a result of PHP [63]. It is plausible, though, that the pain associated with PHP limits function, and as a consequence, muscle size decreases due to atrophy, so this is likely a secondary finding of PHP. However, such a finding helps inform whether muscle atrophy is indeed present with PHP, which may lead to further studies to more rigorously investigate its clinical relevance.

Hyperaemia is the active engorgement of vascular structures and is one of the primary responses to inflammatory stimuli. A meta-analysis of studies that measured hyperaemia in this review was not appropriate due to methodological heterogeneity between studies, however there was evidence from two studies of hyperaemia being more frequent in participants with PHP. Further, the presence of severe hyperaemia was only found in participants with PHP and not in healthy control participants. However, the degree of hyperaemia detected was substantially less in one study [35] than the other study [41], so additional studies are needed to determine with certainty if hyperaemia is associated with PHP, and consequently, whether it is worthwhile evaluating the effectiveness of treatments aimed at optimising the healing process of injured connective tissue structures, such as prolotherapy [64].

Change in the thickness of the plantar heel fat pad has also been studied. People with PHP were found on ultrasound to have a mean loaded fat pad thickness that was 0.97 mm thicker that people without PHP. This is somewhat supported by the mean unloaded fat pad thickness that was 0.48 mm thicker, although this finding was not found to be statistically significant (it was almost statistically significant at p = 0.050). It is currently unknown if these values are clinically important, however a thicker fat pad may be an adaptive response to vertical load; such as prolonged standing, running or a high BMI, a mechanism that has similarly been proposed for calcaneal spur development [65, 66]. If so, using soft orthotic materials or shoe midsoles may dissipate increased force under the heel. Further, contoured orthoses will have a similar effect by decreasing force and plantar pressure under the heel [67, 68].

Several other imaging features relating to bone were also identified. People with PHP were more likely to have plantar calcaneal spurs, bone marrow oedema and increased radioisotope uptake in the calcaneus. In this review, meta-analysis revealed that PHP participants were greater than 5 times more likely to present with a plantar calcaneal spur than control participants, which is slightly lower than the finding in our previous review [14]. A sensitivity analysis of blinded studies found that PHP participants were 12 times more likely to have a plantar calcaneal spur when compared to control participants, which counterintuitively, was higher than the unblinded studies. However, the two blinded studies both used paired heel data, which as stated previously, may affect the independence of the sample and any subsequent statistical analysis. With this in mind, we have elected to focus on the findings from the overall analysis of all studies, which found that people with PHP were 5 times more likely to have a plantar calcaneal spur. Isolated plantar calcaneal spurs are known to frequently co-exist with plantar fascia changes [69], and as such, they are unlikely to represent a discrete clinical manifestation. They are also frequently found in people without PHP [70], are associated with increasing age and obesity, and may be a response to vertical load rather than longitudinal traction at the origin of the fascia at the plantar calcaneus [65], although this is still somewhat under debate [71]. Accordingly, the finding of increased odds of plantar calcaneal spurs in people with PHP is of interest, however it is an association only and unlikely to be the cause of pain. Further, the presence of a plantar calcaneal spur has limited relevance to treatment, unless the spur is fractured, in which case fracture management principles would be necessary [72]. The use of x-rays, therefore, has a limited place in PHP.

The presence of bone marrow oedema was not measured in our previous review [14]. Two studies included in this review measured the presence of bone marrow oedema, although they were not appropriate for meta-analysis. An MRI study that used blinded assessors found over one third of PHP participants had bone marrow oedema in the calcaneus [49]. Interestingly, there was a small sub-group of symptomatic participants with bone marrow oedema who had clinical symptoms of PHP but no abnormalities of the plantar fascia. A moderate association between bone marrow lesions, structural progression, and longitudinal change in pain has been reported in knee osteoarthritis [73]. The foot and ankle has had limited study compared to the knee [74], however bone marrow oedema may present with unique clinical symptoms in PHP such as night pain [75]. The aetiology of bone marrow oedema is still uncertain, but treatment usually involves analgesics and offloading the limb. Further, increased radioisotope uptake in the calcaneum of PHP participants in scintigraphy studies [55, 60] lends support to a subset of people with PHP who have increased metabolic bone turnover within the calcaneus. The exact physiological process for this condition is unclear, but it is likely to be load-related and represents a target for further evaluation to determine its clinical relevance. If such a condition is found to be definitively associated with PHP, then this may represent a fatigue or stress injury of the bone. Another study [47], found that PHP participants also had greater cross-sectional area of the calcaneal tuberosity trabeculae (calcaneal crescent sign), which supports a fatigue or stress injury hypothesis, or at least a response to bone stress. The lack of studies investigating this feature precludes a definitive statement on the relevance of the crescent sign at this stage, although if further studies support this finding, it would be in keeping with a bone stress phenomenon in PHP.

This systematic review was designed to be a comprehensive review of the literature, however its findings should be considered in relation to several limitations. Firstly, it is possible that some appropriate studies may not have been identified and included. As in our previous review [14], studies were only included if they reported medical imaging findings in adult participants with PHP and compared these findings with those from independent control participants who were asymptomatic of PHP. In doing so, 15 studies that did not meet these criteria were excluded and therefore, all imaging features associated with PHP may not have been included in this review. Secondly, there was substantial heterogeneity in most of the meta-analyses and only one-fifth of studies were rated ‘good’ on quality assessment. Over half the studies included bilateral heel data from the same participant, which could have affected the results of our meta-analyses, and a similar proportion did not report where the study sample was recruited from or participants’ BMI, which leads to generalisability concerns. The majority of the studies did not use assessors who were blinded to the status of the participants (i.e. whether or not they had PHP), which could have led to assessor bias. This may have over-estimated the associations or differences we found, however we tried to take a conservative approach to this issue and conducted sensitivity analyses where appropriate. While the extent of these associations changed depending on blinding, what did not change was whether an association existed. The majority of studies also did not report inter- and intra-assessor reliability for imaging observations, which may have affected the accuracy of the imaging observations made. Ideally, assessors should demonstrate both intra- and inter-assessor reliability, which is something future studies should determine prior to data collection. Lastly, some of the meta-analyses included only two studies, and relatively small sample sizes, so the precision of the estimates of the associations for these analyses may be less than ideal. Further studies investigating these associations should improve the precision of these estimates.