Though there is no gold standard for monitoring the progression of knee or hip osteoarthritis (OA), there is value in monitoring the disease[1, 2]. Knowledge of the trajectory of disease progression provides clinicians and patients with benchmarks against which the effectiveness of everyday self-management strategies[3] or clinician-provided interventions can be evaluated[1, 2]. Furthermore, the timing of knee or hip arthroplasty for people with OA may also be informed by capturing significant deterioration in various health domains of those waitlisted for surgery when wait times are protracted. For example, those waitlisted for surgery may be escalated if there is documented evidence of significant decline since first consenting to the procedure.

There are multiple tools available which capture disease severity based on the symptoms and impairments associated with OA. These include tests of physical function and mobility as well as patient-reported surveys. Thoughtful monitoring of the clinical severity of OA in an individual using these tools requires knowledge of what changes measured by these tools can be considered real. In order to arrive at a decision, the clinician must first be cognisant of the minimum change measured by a given tool that is considered to be more than simple measurement error[4]. This minimum change is referred to as the minimal detectable change (MDC)[5–7] or smallest detectable change (SDC)[4, 8, 9] and is mathematically (linearly) related to the error of the measurement. Put simply, the MDC or the SDC reflects the smallest within-person change in score that can be interpreted as real and statistically significant[8]. In terms of clinimetrics, the MDC is a metric for reproducibility (specifically, a measure of agreement), and is determined by performing repeat measurements on patients over a short period of time[8]. The short time interval renders significant clinical change between assessments unlikely[8], and it also avoids the problem of response shift - a change in the meaning or a recalibration of an outcome - if the tool captures a patient-reported outcome[10].

Despite their obvious value in interpreting real change at the level of the individual, several recent reviews of tools used to assess OA and arthroplasty patients suggest that the measurement error and MDC for the tools are under-reported or underexplored[11–16]. This study, therefore, aimed to evaluate the MDC of tools commonly used to assess the symptomatic and functional severity of knee or hip OA. Specifically, using a test-retest design, we determined the MDC for the visual analogue scale for pain (VAS Pain)[1, 17, 18], the timed up-and-go (TUG) test[19], the six-minute walk test (6MWT)[20], the Knee Injury and Osteoarthritis Outcome Score (KOOS)[21] and the Hip Disability and Osteoarthritis Outcome Score (HOOS)[22] in individuals with severe OA. As a secondary aim, we also compared the MDCs for the different tools as the magnitude of error may influence the choice of tool used. The tools included were chosen by a multidisciplinary working party overseeing a State-wide program primarily intended to screen, treat and monitor patients with severe knee or hip OA waitlisted for arthroplasty surgery - the Osteoarthritis Chronic Care Program (OACCP)[23].

Given the complexities and associated burden of quantifying change from one visit to the next[37], one could argue why not simply ask the patient if they have improved or worsened. Though the latter approach may be appealing, in cases where patients may benefit from reporting deterioration - for example, by being escalated to arthroplasty surgery - transparency and quantification of change are required. Further, protracted periods between assessments will undermine both patient and clinician recall, thus, one’s ability to recognise change will be in doubt[1]. Consequently, an objective method for interpreting ‘change’ is required.

The use of the MDC or other indices of error (or reproducibility) to determine thresholds for change is an objective, transparent, simple way to help the clinician monitor change in the individual with OA. Here, we provide the MDC for a range of tools with the 6MWT demonstrating (in a relative sense) the smallest measurement error across the physical and patient-reported tests assessed. For a change in these tools to be considered ‘real’, the patient with knee or hip OA respectively would need to demonstrate a minimum change (at the 95% level of confidence) of at least 79 and 81 m for the 6MWT, 3 cm for VAS Pain, 37% and 41% for TUG (from baseline), 20 and 22 for KOOS and HOOS Pain, 24 and 23 for KOOS and HOOS Symptoms, 21 and 18 for KOOS and HOOS ADL, and 27 and 24 for KOOS and HOOS QOL subscales.

The magnitude of the week-to-week variation and MDCs observed here are generally consistent with others. An earlier study observed that 90% of stable patients with musculoskeletal problems demonstrated a week-to-week change of up to 3-points (27%) in their pain rating on an 11-point numeric rating scale[38]. Kennedy et al.[28] reported the MDC₉₀ for the TUG and 6MWT as 2.49 s and 61.34 m in their combined TKA and THA cohort. Our MDC₉₅ for the KOOS subscales were generally smaller than the 95% LOA (equivalent to the MDC₉₅) for those reported by Roos and Toksvig-Larsen[21] which ranged from 40 (Symptoms) to 60 (Sport and Recreation subscale), whilst our MDC₉₅ for the HOOS subscales were slightly larger than those reported by Ornetti et al.[39] which ranged from 10 to 20. In terms of how the agreement indices of the KOOS and HOOS Pain, Symptom and ADL Function subscales compare with the OKS and OHS, we found that the CV% were similar (16% for both OKS and OHS)[32]. These observations are interesting as it appears the greater specificities afforded by the KOOS and HOOS subscales do not guarantee a smaller measurement error compared to a survey that does not differentiate contributions made by pain and functional impairment. It is noteworthy that Impellizzeri et al.[40] reported a much smaller CV (7%) for the OKS; this is in part explained by the reverse scoring method (12–60, with low scores denoting less pain and impairment) used in their study.

The MDCs are related to the size of the measurement error. Our study design does not allow us to determine whether the error we have observed is due to within-individual inconsistency in interpretation of survey questions or subclinical, random biological fluctuations in the case of timed walk or continuous pain scale measurements. Nevertheless, the fact remains that ‘noise’ in outcome measures – whether they be objectively or subjectively measured - will undermine the capacity to monitor disease progression, thus, knowledge of the MDC of each tool is important for reliable interpretation of disease status. Our study design also does not allow us to determine what changes are clinically relevant. Whilst the MDC₉₀ and MDC₉₅ values provide clinicians with a threshold about which ‘true’ change can be considered to have occurred with considerable confidence, these thresholds do not denote thresholds for clinically important changes. Reference to the minimal clinically important difference (MCID) or minimal important change is expected to assist the determination of whether change is clinically relevant[1, 8]. This notwithstanding, it is unclear what these values are for all these tools and their interpretation is contentious given that the MCID appears to vary according to baseline severity and the scale used to determine it, and may be time-dependent[1]. For now, then, the MDCs provide a robust alternative for interpreting change in the clinic.

We acknowledge the strengths and limitations of our study. We examined a well-defined cohort likely to be representative of patients with severe OA waitlisted for arthroplasty. This contention is supported by the observations that: 1) the age (68 and 65 yrs), BMI (34 and 31), and gender (female, 63 and 54%) profiles of the knee and hip cohorts respectively, reflect those of the entire patient populations waitlisted for hip or knee arthroplasty at the two sites involved (age, 69 and 65 yrs; BMI, 34 and 30; female gender, 68 and 58%, knee and hip cohorts respectively) as per the data each site routinely collects for submission to the State’s arthroplasty registry (Arthoplasty Clinical Outcome Registry for NSW, ACORN), and; 2) the baseline physical and patient-reported characteristics of our cohorts reflect those reported elsewhere[30–32, 39–41]. Our sample size exceeded the minimum recommended sample size for reproducibility studies, we therefore contend the error margins are credible estimates and not unduly influenced by an inadequate sample size. We tested reproducibility under usual care conditions, thus avoiding overly optimistic error estimates. We have provided reproducibility indices of a range of tools commonly used to assess knee and hip OA in the one study allowing comparisons across the tools. In terms of limitations, we relied on participant perception of their stability in their health status and we did not challenge their declarations that they did not change their medication or physical activity levels between the two test days. Our assumption around stability of health status was necessary as there is no known gold standard for assessing stability in OA[1]. We deliberately avoided the arbitrary use of stability in one of the tools, for example VAS joint pain, as the criterion for participant inclusion in the analysis as this assumes superior reproducibility of the chosen criterion over all others. Nevertheless, 90% (64/71) of the knee and 84% (48/61) of the hip cohort demonstrated a test-retest difference of ≤ 2 points in VAS pain (details not shown in Results). Importantly, these changes align with the weekly changes observed in the LEAP Trial in a cohort of patients with OA who were considered stable[42]. Further support for our contention that the participants were stable is found in the observations (results not shown) that a change in one measure was not reliably associated with a change in another, both in terms of magnitude or direction, suggesting that the changes were, by and large, ‘noise’. Regarding unchanged medication and activity levels, participants were aware that their waitlist assessment was to be conducted over two assessments and these would be used by the clinicians to inform their management whilst waiting for surgery. Thus, we contend participants were unlikely to have changed their management knowing that the intention of the assessment (at least from a clinical assessment perspective) was to assess the appropriateness of their current management and to provide a new plan if deemed necessary.

Knowledge of the MDC values for physical performance and patient-reported tests commonly used to monitor the severity of OA is necessary for interpreting change within the individual in the context of daily clinical practice. The 6MWD demonstrated the smallest measurement error and, thus, has the capacity to detect the smallest real change above measurement error, making it (potentially) the preferred measurement tool.