This systematic review with narrative synthesis is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines and was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO); registration no. CRD42018107439 available from https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42018107439 [15].

We included published prospective or cross-sectional experimental and observational studies with adult participants (age > 18 years) with PHP, heel spur syndrome, plantar fasciitis, plantar fasciopathy, plantar fasciosis, painful heel syndrome or calcaneodynia. We excluded studies that were undertaken in populations with differential diagnoses of PHP such as spondyloarthritis, fat-pad atrophy, proximal plantar fibroma, calcaneal stress fracture, insertional Achilles tendinopathy, and non-painful heel spurs. In addition, we excluded retrospective studies (studies in which outcomes of interest were collected before the start of the study). We conducted literature searches in the following databases: PubMed, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science (WoS), CINAHL, and Physiotherapy Evidence Database (PEDro). The search was limited to studies published between January 1st, 1998 and March 4th, 2019, and there were no language restrictions. The search strategy was developed in collaboration with a librarian with expertise in designing such strategies for systematic reviews and was adapted to fit each database. We used the following search terms: plantar fasciitis, heel spur*, heel pain, policeman’s heel, calcaneal spur*, calcaneal pain, plantar fasci*, and plantar aponeuros*.

Two authors (HR and MLP) independently screened the studies that were retrieved from the search using Covidence (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia). If a study was published in a language that the two authors were unable to understand, they would consult another researcher with sufficient language proficiency in that language. After screening titles and abstracts, studies were grouped into: (i) randomised trials, and (ii) non-randomised study designs. As we expected a large number (> 500) of studies to be eligible, we decided a priori to randomly select 50% of the randomised trials stratified by year for full-text screening, with the objective of including a representative sample of trials. Trials that were described as randomised, but which were subsequently found on closer scrutiny to be non-randomised were moved to the pool of non-randomised study designs. After the inclusion of randomised trials, we screened (via full-text) the same number of studies with non-randomised designs to arrive at a final sample of studies that included an equal proportion of randomised trials (50%) and non-randomised study designs (50%). We used a weighted random selection based on the year the included randomised trials were published to match the yearly distribution of the non-randomised study designs. When a study with a non-randomised design was excluded during screening, we included a new study published in the same year as the excluded study and continued this process until the number of included studies with non-randomised designs matched that of the randomised trials. If any disagreements about inclusion arose during screening and consensus could not be reached between HR and MLP, a third author (JLO) was consulted to make the final decision.

Data extraction was performed independently by HR and MLP. Any disagreements were resolved by discussion. Data extraction forms were developed a priori and included: terminology, setting, method of recruitment, country, interventions, eligibility criteria, number of participants, age, height, weight, BMI, symptom duration, pain intensity, and physical activity level. If studies only reported height and mass, we calculated BMI as kg/m2. In studies with multiple groups, we combined the groups and calculated a weighted mean and SD for any participant eligibility criterion that was measured on a continuous scale. If studies reported the median and range, we estimated the mean and SD [16]. In case of missing data or discrepancy between criteria and participant characteristics reported, we contacted the authors via e-mail for clarification. The validity of the diagnosis was assessed independently by HR and MLP and evaluated as either high (based on physical examination and/or imaging), low (based on patients’ self-reported symptoms), or unclear (unspecified). This classification was based on clinical guidelines that recommend physical examination [17].

To analyse the participant eligibility criteria and evaluate the extent of variation (Aim 1), we considered criteria qualitative data and undertook a content analysis in NVivo version 11 (QSR International, Melbourne, Australia). One author (HR) initially divided the criteria into overarching criteria and then the final content analysis was performed together with another author (JLO). As an example, measures of pain were first extracted as worst pain, first-step pain, or average pain over time, forming the overarching criterion ‘pain’. We calculated the number of studies that used criteria within a given overarching criterion, however one study may have had several criteria within the same overarching criterion. For example, several differential diagnoses and comorbidities may have been used as exclusion criteria in a single study, but this study was only counted as one within the overarching criterion called ‘other diagnoses’. To use an overarching framework, we aligned overarching criteria to the International Classification of Functioning, Disability and Health (ICF). Thereby, participant eligibility criteria were classified as being related to either participation, body structures/function, environmental factors, activities, or personal factors [18]. After the content analysis, all authors decided by consensus if there was a high variability in the participant eligibility criteria used. In this process, we considered the number of overarching criteria within the ICF framework as well as the between-study variation of how the same criterion was used. For example, the criterion of ‘pain’ has sub-criteria such as worst pain, first-step pain, and average pain, and these sub-criteria may have different eligibility thresholds on a 0 to 10 cm Visual Analogue Scale (VAS).

To test the association between the criteria of age, BMI, symptom duration and pain intensity with the corresponding participant characteristics (Aim 2), we separated studies into those that specified these as selection criteria from those that did not. Those that pre-specified criteria were further pooled into those that used the same ranges and thresholds for these measures. We also pooled studies that did not include these participant characteristics in their eligibility criteria but used these as a reference value for each of the characteristics. For example, if a study did not use age as a criterion, but still reported participants’ age, this study was pooled into a reference sub-group for age. Based on the sample size of the studies, we calculated a weighted mean of the reported characteristics within each of the sub-groups and compared these between groups. We did not use any inferential statistics as weighted means were used so the common assumptions were not met in our design. Instead, we present the raw data in plots to enable the reader to interpret the data.