Standing balance assessment and prognosis in knee osteoarthritis

Executive assessment

Standing balance assessment can reveal important limitations in knee osteoarthritis, particularly when ordinary walking remains possible and a broad functional scale has reached its ceiling. The most defensible approach is to select a small number of tasks that answer explicit clinical questions: maintaining single-limb support, transferring weight while stepping, reaching toward the edge of support, recovering from a perturbation, or sustaining stability when vision is removed. These tasks impose different mechanical and sensory demands. Their scores should remain separately interpretable rather than being collapsed into an unvalidated knee OA balance score.

For independently ambulant people with mild-to-moderate medial knee OA, the Community Balance and Mobility Scale (CB&M) has useful direct evidence and substantially less ceiling limitation than the Berg Balance Scale (BBS). For a brief assessment, single-leg stance, the Step Test and Functional Reach are practical. The newer Mostafaee study reports direct minimal detectable change (MDC) and anchor-based minimal important change (MIC) for all three; it materially updates the older literature. Its values must nevertheless be matched to the original protocol and patient spectrum. Full-text appraisal shows that its global anchor classified “much better” and “very much better” knee function as improvement, while “slightly better” was grouped with no change. Its reported MICs therefore require that important qualification, as well as protocol matching, before clinical use. [1, 2]

The Star Excursion Balance Test (SEBT) is a useful higher-demand option. The strongest broad clinical study used eight directions, one practice and two measured trials, and the mean normalized composite. Its MDC95 was 8.45 percentage points of leg length. A later end-stage OA study used a different three-direction protocol, more practice and three measured repetitions, and reported much smaller direction-specific MDCs. These estimates concern different tasks and samples; the later result does not supersede the earlier composite threshold. Responsiveness across knee replacement also does not establish nonoperative OA responsiveness. [3, 4]

Force-platform and wearable results require more discrimination than the word “objective” suggests. COP velocity and path length have better reported repeatability than area in the small direct knee OA single-leg study. An IMU study discriminated some severe-OA movement strategies, but its repeated-testing reliability was estimated only in healthy controls. A newer accelerometer study of anticipatory adjustments achieved strong cross-device correlations, yet within-session ICCs were only approximately 0.47–0.56 in the OA group, with unresolved table labels and change-threshold reporting. Neither study validates a generic phone-derived balance or falls score. [5–7]

Prospective falls evidence exists, but it remains insufficient for an individual risk calculator. A 2025 perturbation study followed 24 community-dwelling participants with monthly falls calendars for a year. Its model had an AUC of 0.66 despite selecting among 128 candidate additive combinations. This is an exploratory temporal association with poor discrimination, not an externally validated tool. In contrast, an apparently impressive two-test clinical “prediction rule” classified recalled falls from the preceding year. These designs must not be conflated. [8, 9]

For rehabtools, a useful first implementation would record task completion, side, support, pain, raw performance, trial rule and protocol-matched uncertainty, with clear separation of observed balance, perceived confidence and actual future-fall prognosis. The major opportunity is trustworthy longitudinal measurement. The major risk is converting cross-sectional group separation, a proprietary fall-risk index, or a development-sample cutoff into an apparently precise forecast.

Scope and interpretation framework

The population must remain explicit

This report concerns adult symptomatic and/or radiographic knee OA before joint replacement. It distinguishes clinical OA, radiographically established disease, severe unilateral disease awaiting arthroplasty, and broader cohorts selected for knee pain or risk of OA. Kellgren–Lawrence (KL) grade 1 is doubtful radiographic OA and should not silently be equated with definite KL grade 2 disease. A patient with substantial symptoms and modest radiographic change is also different from an asymptomatic person with structural OA. The compartment involved, alignment, unilateral or bilateral symptoms, body mass, age, activity and use of a walking aid all alter interpretation.

End-stage OA assessed before surgery belongs within this review as a distinct subgroup. Measurements after TKA belong primarily in the companion TKA review. A study that uses preoperative participants for repeatability and postoperative participants for responsiveness supplies two different types of evidence. “Prospective” recruitment does not make a same-day association a prognostic study.

The balance construct is similarly broad. Quiet standing measures regulation of equilibrium within a relatively fixed base of support. Single-leg stance adds substantial load tolerance and support demands. Step and reach tasks assess anticipatory control while moving the body or the other limb. Perturbation recovery examines reactive responses. Composite scales add transfers, locomotion, speed and dual tasking. Knee pain, weakness, range of motion, fear, attention, vestibular function and ankle/hip strategies can all affect performance. A low score identifies a task limitation; it does not uniquely localize its cause to proprioceptive failure at the arthritic knee.

Measurement and prognosis answer different questions

An ICC describes the preservation of relative ranking. It can be high in a heterogeneous group despite clinically substantial within-person error. SEM and MDC/SDC describe error in specified units, under a particular repeat-testing design. MIC describes patient-important change and requires an appropriate anchor. Group responsiveness, usually expressed with a standardized response mean or effect size, does not automatically establish either individual detectability or importance.

For instrument agreement, correlation indicates association but can coexist with constant or proportional bias. A wearable and a camera system may rise and fall together while producing different absolute values. Comparison with a balance scale is convergent validity, not calibration of a COP trajectory against a mechanical reference standard. A body-worn sensor estimates acceleration or model-derived COM movement; a force platform measures ground-reaction-force-derived COP. These quantities are related but not interchangeable.

Actual future falls, recalled prior falls, fear of falling, balance confidence and a laboratory “fall-risk” index are distinct endpoints. A study that discriminates KL grades or a low balance-score category is not forecasting falls. For a proposed risk model, the relevant questions include the follow-up horizon, ascertainment, event count, model selection, internal validation, calibration, incremental value and external validation, rather than only whether a regression coefficient is significant.

Search approach and evidence access

Searches completed on 2 October 2026 combined knee OA, balance or postural terms, and measurement or prognostic terms with an explicit publication-date cutoff. The dated PubMed query yielded 310 records, verified against the official NCBI records. A focused Scopus search used named standing, sway, single-leg, SEBT, CB&M and force-platform terms with reliability, validity, responsiveness, measurement error, prediction, prognosis or falls and yielded 118 distinct records.

This critical narrative review does not claim registration, exhaustive coverage or duplicate independent screening. Detailed appraisal prioritized original measurement studies, prospective outcomes, contradictory findings and implementation-relevant technologies. Complete articles were examined for CB&M, both SEBT studies, IMU single-leg measurement, accelerometer anticipatory adjustment, recent posturography and the prospective perturbation study. The original Mostafaee article and Takacs COP article, including their methods and tables, support the anchor interpretation and the number of patients analyzed. Some older association studies remain abstract-only; their unobserved details are not reconstructed from secondary citations. The bibliography identifies source access and remaining limitations.

Clinical standing and functional balance tests

What the older reviews establish

Hatfield and colleagues' review showed that people with knee OA often perform worse than comparison participants on clinical standing-balance tasks. Its eight included studies covered the Step Test, BBS, single-leg stance, Functional Reach, tandem stance and CB&M. The pooled standardized difference was −1.64 (95% CI −2.58 to −0.69), but heterogeneous tasks and selected samples make this unsuitable as a universal severity scale. Crucially, no included study directly compared clinical balance across radiographic severity levels. The review also excluded people requiring a walking aid and studies without an appropriate comparison group. Its conclusions therefore describe selected ambulant populations rather than all patients presenting for rehabilitation. [10]

The later review by French and colleagues explicitly considered psychometric properties and the domains of postural control assessed. It is useful for mapping an incomplete test portfolio, particularly the difference between maintaining, anticipating and reacting to instability. Neither review provides a single OA-specific fall-probability cutoff. Newer original studies are necessary when judging error and meaningful change. [11]

Single leg stance and the importance of the stopping rule

Single-leg stance is inexpensive and clinically understandable, but “time standing on one leg” is not a sufficiently complete protocol. The symptomatic versus preferred limb, eyes open or closed, footwear, hand position, stance-knee flexion, permitted trunk motion, number of attempts and the maximum time all matter. The patient may stop because of pain or fear before equilibrium is actually lost. A 30-second cap produces a different distribution and ceiling from a 90-second cap; best-of-three and mean-of-three scores likewise have different errors.

The direct CB&M validation cohort provides an informative 90-second example. Twenty of 25 medial-OA participants repeated testing after a mean eight days (range 4–14). The single-leg stance ICC was 0.91 (95% CI 0.78–0.97), but SEM was 11.7 seconds (95% CI 9.23–16.06), and MDC95 was 32.5 seconds. Mean scores were 44.4 and 46.4 seconds, with SDs about 33.5 seconds. The high ICC coexisted with large absolute uncertainty. It cannot justify interpreting a five-second improvement as real change under that protocol. [1] (original Table 4)

Mostafaee's study used the best of three single-leg trials on the affected limb, stopping for foot contact or excessive trunk motion. The article does not specify a maximum duration, vision condition or arm-position rule. Among 70 participants who reported no knee-related change during the one-week retest interval, ICC(2,1) was 0.79 (95% CI 0.68–0.87), SEM 2.58 seconds and MDC95 7.15 seconds. Mean scores were 9.98 and 10.97 seconds. This substantially lower-performing sample and different protocol should be distinguished from the older 90-second-cap study; neither estimate should be averaged or applied to the other task. [2] (original Table 2)

The same study's reported 13.10-second MIC was derived after four weeks of physiotherapy from a global rating of knee function. The anchor counted “much better” or “very much better” as improved and “no change” or “slightly better” as unchanged. It therefore separates a substantial perceived improvement from no or slight improvement, rather than directly locating the smallest patient-perceived benefit. This qualification is essential when interpreting the published label MIC. [2]

Cross-sectional findings also require care. Hunt and colleagues investigated correlates of single-leg stance in medial OA; the direction or strength of relationships with radiographic severity can reflect adaptation, alignment and selection. An independent-ambulator sample aged at least 80 years is particularly selected and cannot define a normal value for a frail, assisted patient. Single-leg inability should be recorded as inability, including the reason and assistance required, rather than as missing-at-random or as “zero balance.” [12, 13]

Step Test and Functional Reach

The Step Test used in the older knee OA literature has the patient maintain stance on the study limb while the opposite foot steps onto and off a 15-cm step repeatedly for 15 seconds. It therefore tests the supporting limb and repeated weight-transfer control rather than simply the moving limb. Step height, distance from the step, hand support and what constitutes a completed repetition need standardization. The task can be limited by pain, hip flexion, speed and confidence as well as balance. [10]

Functional Reach measures a voluntary reach displacement under a specified support condition. It permits different ankle, hip and trunk strategies, making it a capacity measure with limited mechanistic specificity. Reach distance is not a direct measurement of the COM margin of stability and does not test an unexpected recovery step. The same arm, starting posture, foot position and heel-lift rule should be maintained between visits.

Mostafaee's full protocol used the 15-cm, 15-second Step Test once, barefoot and without hand support, following two or three practice steps. In bilateral disease, the more radiographically severe knee was the supporting limb. On loss of balance, the trial stopped and the repetitions already completed were retained. Functional Reach was the mean of three attempts next to a shoulder-height yardstick without moving the feet. The stable-retest Step Test ICC was 0.82 (95% CI 0.69–0.89), SEM 0.98 steps and MDC95 2.71 steps; Functional Reach ICC was 0.76 (0.65–0.85), SEM 1.77 cm and MDC95 4.90 cm. The lower confidence bounds warrant more nuance than the authors' point-estimate label “excellent.” [2]

Baseline correlations with TUG were −0.74, −0.66 and −0.61 for Step, stance and reach respectively, versus −0.57, −0.54 and −0.49 with the WOMAC total score. This meets the prespecified expectation of stronger relationships with observed mobility than with self-reported disease burden. For responsiveness, 90 participants attended 12 physiotherapy sessions across four weeks; the other ten planned to receive PRP in addition and were excluded from this analysis, rather than being lost to follow-up. Change correlations with the global anchor were 0.57, 0.53 and 0.51, supportive but not evidence that the anchor measures balance alone.

ROC AUCs were 0.86 (95% CI 0.77–0.94), 0.81 (0.72–0.89) and 0.78 (0.69–0.88). Youden-selected thresholds were 4.5 steps, 13.10 seconds and 5.5 cm, with sensitivity/specificity 0.83/0.86, 0.70/0.77 and 0.77/0.77. These are direct, useful anchor-based data, but the thresholds were selected and evaluated in the same sample and are conditioned on the “much better” anchor definition. No MIC confidence intervals, responder counts or explicit handling of deterioration were found. For integer Step Test counts, 4.5 is a classification boundary separating changes of four and five steps, not a physically observed half-step gain. For reach, the threshold is only 0.60 cm above MDC95, making threshold uncertainty and ruler precision consequential.

The sample's mean age was 62.3 years and 57% had bilateral disease. Table 1 lists 12 KL1, 42 KL2 and 39 KL3 cases, totaling 93 of 100; the remaining seven grades are not identified. The inclusion criteria exclude arthroplasty only in the preceding six months, so a replacement-naïve sample cannot be assumed. These reporting limits, the knee-function rather than balance-specific anchor, and the unspecified single-leg cap prevent universal transfer while preserving the study's contribution to short-term clinical assessment. [2] (original pages 1704–1707)

Table 1 Direct knee OA estimates for balance measurement and reported important change

Mostafaee 2025, Tables 2 and 4. Reliability n=70; responsiveness n=90 after four weeks. Anchor improvement requires much or very much better overall knee function; slightly better is grouped with no change. MIC uncertainty and responder counts not reported. Values are not general falls cutoffs. [2]

Community Balance and Mobility Scale and Berg Balance Scale

Takacs and colleagues studied 25 people with radiographic medial OA and 25 sex-matched controls. Most OA cases were KL2 (14) or KL3 (9), with only two KL4 cases. The CB&M contains 13 challenging tasks, some bilateral, with a maximum score of 96. It includes turning or looking while walking, unilateral support, running, step-ups and stair descent. It requires approximately 15 minutes, an 8-m walkway and simple equipment; it therefore assesses advanced functional balance and mobility rather than isolated quiet standing. [1]

Mean CB&M scores were approximately 71 in OA and 85 in controls, and absolute correlations with comparison balance/mobility tests were 0.52–0.74 in OA; the TUG correlation was r=−0.74. In contrast, the BBS was at its maximum in 64% of OA participants and 92% of controls. This is clinically important: a high BBS score in an independently ambulant patient may leave higher-level deficits unresolved.

For the 20-person retest subgroup, the CB&M ICC(2,1), absolute agreement, was 0.95 (95% CI 0.70–0.99). Original Table 4 reports MDC95 of 9.4 points; the discussion rounds this to 10, a sensible whole-score operational interpretation. The BBS ICC was 0.59 (−0.07 to 0.84) in its restricted range, with MDC95 2.3 points, rounded to three in prose. These figures are not evidence that BBS is intrinsically poor across all knee OA severity. Range restriction and ceiling effects limit its usefulness in this particular ambulant sample. The authors appropriately call for future falls validation; no future falls endpoint was assessed. [1]

Star Excursion Balance Test

Kanko and colleagues provide comparatively strong direct clinical measurement evidence: 74 patients met clinical OA criteria, 37 also had simultaneous 3D motion capture, and 66 returned after 12 weeks. Participants unable to stand on one limb for five seconds were excluded. The eight-direction test was performed barefoot, with hands on hips, the stance foot fixed and the stance knee aligned over the toe. There was one practice in each direction and two recorded trials, averaged. Reach was normalized to the ASIS-to-medial-malleolus leg length. [3]

The affected-limb composite ICC was 0.94 (95% CI 0.91–0.96) for centimetres and 0.93 (0.89–0.96) after normalization. SEM was 2.68 cm or 3.05 percentage points of leg length; MDC95 was 7.44 cm or 8.45 percentage points. Observer and camera reach measures were highly correlated (r≥0.96), but the analysis emphasized correlation rather than a complete interchangeability evaluation with bias and limits of agreement.

After 12 weeks, mean normalized reach improved by 4.46 percentage points (95% CI 2.97–5.95), with SRM 0.74. The average change was smaller than the individual MDC95, an entirely possible combination: a group can show a responsive mean change while many individuals cannot be distinguished from error with high confidence. Change correlated only weakly with walking and pain and not significantly with KOOS subscales. These findings support sensitivity to the tested neuromuscular task, not a substitute measure of every aspect of knee function. Adherence to the home programme was not measured, so the study should not be repurposed as a definitive treatment-effectiveness trial.

The 2024 study by Bin Sheeha and colleagues evaluated a different, three-direction SEBT in 35 people with unilateral KL4 OA awaiting TKA. Four familiarization attempts preceded three measured trials, averaged and normalized. Preoperative one-week ICCs were exceptionally high: 0.998 anterior, 0.998 posteromedial and 0.993 posterolateral. Table 1 reports SEM/MDC of 0.42/1.18, 0.37/1.02 and 0.68/1.89 percentage points of leg length, respectively. The protocol, severe unilateral sample and extensive practice prevent direct substitution for Kanko's eight-direction composite. [4]

The same study showed large pre-to-post-TKA changes, but changes in SEBT did not correlate significantly with changes in walking, chair stand, stairs or Oxford Knee Score. Interpretation should retain those null results. Several reporting anomalies also limit precision: the recruitment arithmetic does not reconcile as written (94 invited minus 56 ineligible and 16 declining does not yield 38), Table 2 labels Pearson coefficients despite the methods specifying Spearman, and a posterolateral six-to-twelve-month difference in Table 4 is incompatible with Table 3 means. The study's statement that MDC establishes “clinical impact” conflates detectability with importance. No anchor-based MIC is established by that analysis.

Instrumented standing and wearable assessments

Force platform centre of pressure

The complete Takacs COP article makes an important correction to its abstract-level sample description. Twenty-five participants volunteered, but five who could not sustain ten seconds of single-leg stance were excluded from analysis. Those five were about ten years older on average than the 20 analyzed participants. Fifteen of 25 needed more than three attempts to obtain the three required trials, with at most six attempts permitted. The analyzed group comprised ten KL2, nine KL3 and one KL4 case. Thus, the error estimates apply to selected successful completers, not the frailest or most unstable patients. [5]

Participants stood on the more symptomatic limb, with collection beginning once stable. Sampling was 50 Hz, the three ten-second trials were averaged, and the mean interval was 8.1 days. Path-length ICC(2,1) was 0.87 (95% CI 0.70–0.95); SEM was 7.28 cm and MDC95 20.19 cm. The mean path was approximately 63 cm, so a sizeable change was required to exceed estimated individual error. Velocity also had ICC 0.87, while area was 0.54 (0.13–0.79). The study shares a laboratory and recruitment approach with the CB&M report, but sex and age distributions differ, so these are not identical cohorts. Any partial participant overlap is unreported and should not be asserted.

The original tables contain unresolved numerical and unit problems. A 63.82-cm path in ten seconds corresponds to approximately 0.064 m/s, whereas Table 1 labels its velocity as 0.64 m/s. Table 2's area SEM of 8.38 and mean area 14.20 do not reproduce its displayed SEM fraction of 0.34, and the text's velocity percentage differs from the table. Consequently, the velocity and area change thresholds are not adopted for implementation. Path length's 7.28-cm SEM and 20.19-cm MDC are arithmetically consistent, but remain specific to ten seconds at 50 Hz. It would be unsafe to silently repair the other published values or transfer laboratory error to a phone or a different force board. [5] (original Tables 1–2)

During a fixed-duration recording, mean COP velocity and total path length encode nearly the same information because velocity is path divided by duration. Treating both as independent biomarkers can inflate a feature set without adding a new construct. Area, directional dispersion and frequency features require explicit processing choices. Foot position, anthropometry, filtering, recording duration and failed trials can dominate apparently small OA-control differences. COP area is particularly vulnerable to outlying excursions and short observation windows.

Posturographic case-control studies show group differences, but their applicability varies. A study that separates early gonarthrosis from controls establishes discrimination within that sample, not diagnostic specificity across pain, vestibular disease, neuropathy or fear. A force-plate classifier separating mild and more severe KL strata also predicts radiographic categories rather than future instability. These studies can generate measurement hypotheses without supplying a clinically validated abnormality boundary. [14–16]

A recent 2026 study of 45 KL2–3 participants and 45 controls combined handheld strength, 30-second double-leg posturography and a 30-second-capped single-leg test. It found greater eyes-closed sway area in OA, 3.41 versus 2.25 cm², and a concurrent association between quadriceps force and sway. This adds a directly relevant contemporary sample but no original repeatability, MIC, prospective falls or calibrated prediction evidence. Its suggested 90% limb-symmetry goal is imported from other musculoskeletal settings and was not validated against future falls in this study. The report's adjusted coefficients also need caution: the listed unstandardized quadriceps coefficient of −0.024, predictor SD 42.18 N and outcome SD 0.89 cm² do not reproduce its standardized coefficient of −0.41 under ordinary standardized linear regression. This unexplained discrepancy argues against copying the equation into software. [17]

Inertial sensors during single limb support

Van der Straaten and colleagues studied 19 severe unilateral OA cases (18 KL4 and one KL3) and 12 controls for validity. Both systems detected some OA-associated changes in trunk lean, pelvic movement and COM displacement, but the camera system identified additional knee differences that the IMU system did not. For mediolateral COM displacement, waveform CMC was 0.72 and RMSE 0.008 m; CMC was not computable for 24% of recordings even after offset correction. Hip angle interpretation was affected by pelvic orientation offsets. [6]

The decisive applicability limitation is that reliability and agreement were examined in a separate sample of 20 healthy participants, not repeatedly tested OA patients. Between-session and between-operator results used the mean of four repetitions, whereas within-session results concerned single measurements. The authors explicitly chose healthy participants to avoid patient movement variability, but that variability is part of clinical OA measurement. Healthy-control MDCs therefore cannot be relabelled as OA change thresholds. The study is useful technology feasibility and discriminant evidence, not proof of reliable home monitoring in end-stage OA.

Anticipatory postural adjustments before a step

Oliveira and colleagues compared an L5 accelerometer at 100 Hz with video-derived L5 acceleration at 120 Hz in 25 OA participants and ten controls. A cued right-foot step was repeated ten times; six successful trials were used for reliability because that was the minimum achieved by all participants. A jump synchronized devices, which itself limits feasibility for some painful or unsteady patients. Heel-marker lift-off defined step onset, so the tested method was not a completely camera-free clinical workflow. [7]

Cross-device OA correlations were 0.98 for anticipatory latency and 0.80 for amplitude. These promising correlations coexist with only moderate repeated-trial ICCs: approximately 0.47 and 0.561 for accelerometer variables in the reported table. The paper's table swaps amplitude/latency unit labels, radiographic-grade counts total 29 despite a stated OA sample of 25, and its MDC description and listed SEM-to-MDC ratios are not clearly consistent with an MDC95 calculation. Accordingly, the published MDC values should not be copied into a clinical threshold field without correction or clarification. The study supports further validation of accessible APA measurement and documents smaller anticipatory amplitudes in OA, while neither establishing between-day monitoring nor predicting falls.

Prognosis and future falls

A true prospective perturbation study

Downie, Levinger and Begg recruited 24 community-dwelling, self-ambulatory adults aged at least 60 with clinically diagnosed knee OA. Following a harness-supported forward lean, a tether release required balance recovery. The starting load was 20% body weight on the overhead transducer. The first recorded trial after two familiarization attempts was analyzed at initial recovery-foot contact, regardless of whether one or several steps were needed. The task requires a specialized laboratory and safety harness. [8]

Twelve monthly calendars prospectively recorded falls, with telephone discussion of reported incidents. Sixteen participants fell and eight did not; 60 actual falls were recorded. Higher forward COM velocity, knee moment and sex entered the selected model. The authors screened candidate variables at p≤0.10 and considered 128 additive models from seven variables. They describe leave-one-out validation, but the retrieved article does not establish that predictor screening and model selection were repeated inside each fold. Validation after selecting predictors on the complete dataset would underestimate optimism.

The reported AUC was 0.66, and the model explained 21.1% of deviance. These represent limited discrimination in a tiny convenience sample, not precise risk estimation. No AUC confidence interval, external calibration, decision-curve analysis or comparison with a prior-falls baseline was found in the recovered article body; all figures and tables were not independently inspected, so article-wide absence has not been established. The apparent sex-specific patterns are especially unstable: the abstract's female percentages imply eight women in total, whereas the contour-plot discussion refers to 14 female points. That unresolved inconsistency prevents confident subgroup interpretation. The reported high predicted probabilities at selected COM/knee-moment combinations are not subgroup AUCs, despite the paper's use of “excellent” and “outstanding discrimination” language for individual predictions.

The useful conclusion is that reactive biomechanics can be studied against actual future falls, and COM regulation deserves further investigation. The study cannot justify a sex-specific clinical cutoff or application to routine quiet-standing sway. It should be cited alongside its poor overall model performance, not solely its positive title.

Retrospective fall classification

Amano and Suzuki's 81-person medial-OA study derived a two-test rule from one-leg stance and five-times sit-to-stand. Cutoffs were 5.3 and 7.9 seconds; both positive tests yielded a reported positive likelihood ratio of 17.8 and positive predictive value of 88.2%. The outcome was a self-reported fall during the preceding year, measured in the same cross-sectional study. Thus the result distinguishes people with recalled previous falls under the study protocol. It does not estimate a future 12-month probability for a new patient. Data-derived cutoffs and predictor selection in the same small sample additionally require independent validation. [9]

Fear-of-falling and balance-confidence scales can add useful patient experience, but they should not be relabelled as objective balance or used interchangeably with fall counts. A patient may avoid activity and have few falls despite marked impairment; another may have good capacity but more exposure to challenging environments. This exposure issue complicates any direct conversion from a laboratory task to community falls.

Function decline and perioperative outcomes

The Observational Arthritis Study in Seniors provides an important longitudinal boundary. Among 480 adults aged at least 65 with chronic knee pain, baseline balance was associated with current disability and performance after strength adjustment, but predicted later decline only in the car-transfer task. It did not predict decline in walking, stair performance or self-reported disability. The sample was selected for knee pain, not exclusively established radiographic OA. [18]

A related 30-month analysis reported declines in strength and dynamic balance, with stronger baseline knees associated with a smaller expected decline in balance. These reports share the OASIS cohort and are not independent replications. They support a relationship between strength and maintaining challenging balance capacity, while leaving the direction and causal mechanism unresolved. [19]

Falls monitored before and after TKA also offer useful context. However, surgery changes pain, mobility exposure, strength and confidence, so postoperative outcomes are a different prediction target from the natural course of nonoperative OA. Prospective monthly surveillance is preferable to comparing annual retrospective recall, but neither design should be transferred into a general OA falls calculator without population- and time-specific validation. [20]

Practical assessment and implementation

A staged selection approach

For a patient with marked pain, support needs or uncertain safety, begin with history of falls and near-falls, ordinary supported stance and observed transfers. Record the assistance needed and why more demanding tasks were not attempted. A test whose validation sample excluded walking-aid users should not be used to interpret that person's inability as a calibrated percentile.

For an independent community ambulator, a brief single-leg stance plus a standardized Step Test or reach task can capture complementary limitations. Choose the limb deliberately, rather than automatically taking a dominant-limb or “best” result. If the BBS is near its ceiling but the patient reports difficulty turning, carrying, stepping or navigating the community, consider a more challenging functional assessment such as CB&M.

SEBT can add information for patients able to tolerate single-limb loading and understand the movement rules. Use either the eight-direction or a specified three-direction protocol consistently. Do not merge their MDC values, omit practice, switch shoes or change between a best and mean score while claiming comparable longitudinal measurement. Force-platform or IMU testing is justified when it answers a specific additional question about sensory conditions, movement strategy or mechanics that cannot be resolved adequately by observation.

Table 2 Select the assessment for the clinical question

Practical synthesis of the evidence appraised in this report. Task, population and protocol determine interpretation.

Required protocol record

The minimum useful record includes diagnosis and severity; affected and tested side; bilateral symptoms; pain immediately before and after; recent analgesia or acute symptom change; footwear and aids; support surface; vision and gaze target; foot position; arm position; instructions; familiarization; number and duration of trials; best/mean/composite rule; stopping and failure definitions; assistance; and actual completed exposure. Instrumented records additionally require device, calibration, sampling, filtering, sensor position, coordinate definitions and algorithm version.

Preserve both raw limb scores. A symmetry ratio can improve because the better limb worsens, and bilateral OA can produce a deceptively symmetric result. An unsafe or incomplete test deserves a valid coded outcome, rather than imputed zero sway or deletion from the dataset. For video, retain separate fields for timing accuracy, scoring reliability, remote supervision and patient feasibility; a valid stopwatch test does not automatically validate automatic video scoring.

What a tool can and cannot conclude

A descriptive tool can show the observed change and the uncertainty estimate from a directly matched protocol, while clearly identifying the source and confidence level. It can distinguish “change exceeds reported error” from “change exceeds a study's patient-important threshold.” Where protocol match or source access is incomplete, display the raw change without a categorical recovery verdict.

The current evidence does not support a universal knee OA balance age, a generalized falls percentage from sway, or a single traffic-light threshold derived by blending retrospective fall classification with prospective studies. A future prediction product would need a defined horizon and endpoint, prospective falls capture, adequate events, prespecified predictors, external validation and calibration in the intended severity and use setting. Even then, utility should be compared with ordinary clinical information rather than assumed from device complexity.

Conclusions

Clinical and instrumented standing-balance assessment can contribute meaningful information in knee OA when protocol and population are explicit. The most mature uses are describing task limitations and monitoring carefully repeated performance. CB&M addresses higher-level limitations that BBS can miss; SEBT adds challenging multidirectional control; short standing tests are practical but strongly protocol dependent. Force-platform and wearable variables offer mechanistic detail with less complete evidence for individual change and prognosis.

The evidence should be presented with its negative findings and boundaries: high reliability can coexist with large error, group responsiveness can be smaller than individual MDC, strong device correlation can coexist with only moderate repeatability, and a statistically selected future-falls model can still discriminate poorly. These distinctions are essential for a rehabilitation tool that helps clinicians interpret a patient rather than merely report a number.

Primary study characteristics

Primary studies supporting the narrative are grouped by measurement or prognostic question. Any consensus recommendation is explicitly identified. Population, protocol, endpoint and source access constrain interpretation. The linked bibliography identifies source-access limitations. Related publications from one cohort are not independent replications.

Table 3 Clinical balance tests and their measurement properties

Table 4 Instrumented balance and postural control

Table 5 Falls and functional outcomes

References

References are numbered in first citation order. Study specific source descriptions identify the material examined and do not constitute a study quality rating. Links identify the original publication or the explicitly named primary source version.

1. Takacs J, Garland SJ, Carpenter MG, Hunt MA. Validity and reliability of the community balance and mobility scale in individuals with knee osteoarthritis. Physical therapy. 2014;94(6):866-74. DOI 10.2522/ptj.20130385 Source examined: Complete article and original Tables 1–4.

Source note: SRC-e86a265861e6 Takacs J 2014

2. Mostafaee N, Pirayeh N, Moosavi SS. Reliability, validity, responsiveness and minimal important changes of common clinical standing balance tests in individuals with knee osteoarthritis. Physiotherapy theory and practice. 2025;41(8):1703-1711. DOI 10.1080/09593985.2024.2445143 Source examined: Complete original article and Tables 1–5.

Source note: SRC-52daaaedf4b1 Mostafaee N 2025

3. Kanko LE, Birmingham TB, Bryant DM, Gillanders K, Lemmon K, Chan R, et al. The star excursion balance test is a reliable and valid outcome measure for patients with knee osteoarthritis. Osteoarthritis and cartilage. 2019;27(4):580-585. DOI 10.1016/j.joca.2018.11.012 Source examined: complete article body; numeric tables unavailable; aggregate results available in the article text.

Source note: SRC-b49834f25f63 Kanko LE 2019

4. Bin Sheeha B, Bin Nasser A, Williams A, Granat M, Johnson DS, Althomali OW, et al. Reliability of the Star Excursion Balance Test with End-Stage Knee Osteoarthritis Patients and Its Responsiveness Following Total Knee Arthroplasty. Journal of clinical medicine. 2024;13(21). DOI 10.3390/jcm13216479 Source examined: complete article body and official HTML tables; formula MathML checked.

Source note: SRC-00d351b30ecb Bin Sheeha B 2024

5. Takacs J, Carpenter MG, Garland SJ, Hunt MA. Test re-test reliability of centre of pressure measures during standing balance in individuals with knee osteoarthritis. Gait & posture. 2014;40(1):270-3. DOI 10.1016/j.gaitpost.2014.03.016 Source examined: Complete publisher HTML including original tables; source inconsistencies retained.

Source note: SRC-cad8cfed0af1 Takacs J 2014

6. van der Straaten R, Wesseling M, Jonkers I, Vanwanseele B, Bruijnes AKBD, Malcorps J, et al. Discriminant validity of 3D joint kinematics and centre of mass displacement measured by inertial sensor technology during the unipodal stance task. PloS one. 2020;15(5):e0232513. DOI 10.1371/journal.pone.0232513 Source examined: complete article body and official HTML tables.

Source note: SRC-91a8b2881858 van der Straaten R 2020

7. Oliveira LKR, Marques AP, Igarashi Y, Andrade KFA, Souza GS, Callegari B. Wearable-based assessment of anticipatory postural adjustments during step initiation in patients with knee osteoarthritis. PloS one. 2023;18(8):e0289588. DOI 10.1371/journal.pone.0289588 Source examined: complete article body and official HTML tables.

Source note: SRC-a713aa69a4f5 Oliveira LKR 2023

8. Downie C, Levinger P, Begg RK. Predicting falls from biomechanical response to balance perturbation in older adults with knee osteoarthritis, an exploratory study. Advanced Exercise and Health Science. 2025;2(2):137-142. DOI 10.1016/j.aehs.2025.03.003 Source examined: complete article body; full tables and figures unavailable.

Source note: SRC-d5ab4ea74dd0 Downie C 2025

9. Amano T, Suzuki N. Derivation of a clinical prediction rule to determine fall risk in community-dwelling individuals with knee osteoarthritis: a cross-sectional study. Archives of osteoporosis. 2019;14(1):90. DOI 10.1007/s11657-019-0641-y Source examined: abstract only.

Source note: SRC-15c44749724e Amano T 2019

10. Hatfield GL, Morrison A, Wenman M, Hammond CA, Hunt MA. Clinical tests of standing balance in the knee osteoarthritis population: Systematic review and meta-analysis. Physical therapy. 2016;96(3):324-37. DOI 10.2522/ptj.20150025 Source examined: review full HTML.

Source note: SRC-bfc6d36675e1 Hatfield GL 2016

11. French HP, Hager CK, Venience A, Fagan R, Meldrum D. Psychometric properties and domains of postural control tests for individuals with knee osteoarthritis: a systematic review. International journal of rehabilitation research. Internationale Zeitschrift fur Rehabilitationsforschung. Revue internationale de recherches de readaptation. 2020;43(2):102-115. DOI 10.1097/mrr.0000000000000403 Source examined: abstract only.

Source note: SRC-5a424615ba73 French HP 2020

12. Hunt MA, McManus FJ, Hinman RS, Bennell KL. Predictors of single-leg standing balance in individuals with medial knee osteoarthritis. Arthritis care & research. 2010;62(4):496-500. DOI 10.1002/acr.20046 Source examined: abstract only.

Source note: SRC-96cb7b82a696 Hunt MA 2010

13. Ishii Y, Noguchi H, Sato J, Ishii H, Ishii R, Toyabe SI. Association of knee osteoarthritis grade with one-leg standing balance and quadriceps strength in male independent ambulators aged ≥80 years. Journal of orthopaedics. 2020;21:79-83. DOI 10.1016/j.jor.2020.03.013 Source examined: abstract only.

Source note: SRC-b182cc023ac7 Ishii Y 2020

14. Hassan BS, Mockett S, Doherty M. Static postural sway, proprioception, and maximal voluntary quadriceps contraction in patients with knee osteoarthritis and normal control subjects. Annals of the rheumatic diseases. 2001;60(6):612-8. DOI 10.1136/ard.60.6.612 Source examined: abstract only.

Source note: SRC-7efe2f4eefb3 Hassan BS 2001

15. Pirayeh N, Shaterzadeh-Yazdi MJ, Negahban H, Mehravar M, Mostafaee N, Saki-Malehi A. Examining the diagnostic accuracy of static postural stability measures in differentiating among knee osteoarthritis patients with mild and moderate to severe radiographic signs. Gait & posture. 2018;64:1-6. DOI 10.1016/j.gaitpost.2018.04.049 Source examined: abstract only.

Source note: SRC-6c305240e048 Pirayeh N 2018

16. Kostro AM, Augustynik A, Kuryliszyn-Moskal A, Jamiołkowski J, Pocienè M, Dzięcioł-Anikiej Z. Significance of Selected Posturographic Methods in Diagnosis of Balance Disorders in Patients with Early-Stage Gonarthrosis. Journal of Clinical Medicine. 2024;13(11):3298. DOI 10.3390/jcm13113298 Source examined: complete article body.

Source note: SRC-a904af192b2c Kostro AM 2024

17. Alshahrani A, Reddy RS. Clinically feasible biomechanical assessment of lower limb strength and postural stability in adults aged 50-75 years with moderate knee osteoarthritis: implications for technology-enabled healthy aging. Frontiers in bioengineering and biotechnology. 2026;14:1797065. DOI 10.3389/fbioe.2026.1797065 Source examined: complete article body and tables.

Source note: SRC-e2371eb32366 Alshahrani A 2026

18. Marsh AP, Rejeski WJ, Lang W, Miller ME, Messier SP. Baseline balance and functional decline in older adults with knee pain: the Observational Arthritis Study in Seniors. Journal of the American Geriatrics Society. 2003;51(3):331-9. DOI 10.1046/j.1532-5415.2003.51106.x Source examined: abstract only.

Source note: SRC-c752a83cdc4d Marsh AP 2003

19. Messier SP, Glasser JL, Ettinger WH, Craven TE, Miller ME. Declines in strength and balance in older adults with chronic knee pain: a 30-month longitudinal, observational study. Arthritis and rheumatism. 2002;47(2):141-8. DOI 10.1002/art.10339 Source examined: abstract only.

Source note: SRC-5d8f3f4ed2a2 Messier SP 2002

20. Swinkels A, Newman JH, Allain TJ. A prospective observational study of falling before and after knee replacement surgery. Age and ageing. 2009;38(2):175-81. DOI 10.1093/ageing/afn229 Source examined: abstract only.

Source note: SRC-a97318d926fb Swinkels A 2009