Gait assessment and prognosis after total knee arthroplasty

Main conclusions

Walking recovery after total knee arthroplasty (TKA) cannot be represented adequately by one speed, one questionnaire or one symmetry score. A short standardized walk is a practical measure of walking capacity; a six-minute walk adds sustained walking demand; the Timed Up and Go (TUG) adds rising, turning and sitting; laboratory gait analysis describes how the person walks; and free-living sensors describe walking in an environment that the patient partly selects. These measures overlap, but none is a substitute for all the others. A person can walk faster while continuing to unload the operated knee, and can improve clinic performance without increasing habitual activity.

For routine monitoring, the strongest practical approach is a reproducible short-walk protocol, a record of assistance and walking aids, a complementary endurance or complex-mobility task when indicated, and explicit documentation of pain and postoperative timing. The four-metre fastest-safe walk has useful direct TKA measurement evidence from the preoperative period through one year, but its published error estimates came from consecutive trials within a visit. They do not establish between-day error in rapidly changing postoperative patients. Six-minute walking remains useful when later recovery or endurance is the question. TUG is valuable as composite mobility; it should not be relabelled pure walking speed or a validated postoperative falls classifier. [1–4]

The numerical evidence requires particular care. Several widely cited TKA thresholds are same-day detectable-change estimates, mixed hip/knee estimates, distribution-derived quantities labelled “MCID, ” or preliminary anchor-based values with modest discrimination. A 2015 reliability paper has contradictory table and prose values. A 2026 digital-gait paper has a major discrepancy between its abstract and retrieved results regarding how many patients attained the gait-speed threshold. These sources can inform research interpretation, but the disputed values should not drive automated patient labels. [5–10]

There is genuine prospective evidence: early fast gait speed contributes to a six-month walking-limitation model; early postoperative function and contralateral strength relate to one- and two-year outcomes; and a neighbours-based TUG model has temporally separated testing. Nevertheless, walking limitation, future TUG time, knee flexion excursion and daily step counts are different outcomes. An association with one does not validate predictions of the others. [11–17]

Scope and evidence approach

The principal population is adults undergoing primary elective unilateral TKA for knee osteoarthritis. Evidence from simultaneous bilateral TKA, mixed THA/TKA samples, patients awaiting replacement, revision surgery, unicompartmental replacement and other surgical indications is identified separately. The contralateral knee is not assumed normal: pain, strength, previous replacement and planned surgery on that side can change both observed gait and the interpretation of an operated-to-contralateral ratio.

Time is recorded relative to surgery, rather than using an imprecise label such as “postoperative.” Useful clinical groupings are preoperative assessment, inpatient or immediate recovery, early outpatient rehabilitation, approximately three to six months, and twelve months or later. These are organizational categories rather than universal biological stages. A preoperative measurement that forecasts a later outcome is prognostic evidence; a correlation among variables measured at the same postoperative visit is concurrent evidence; and change across recovery is longitudinal descriptive evidence unless a temporally ordered prediction analysis was performed.

This is a critical narrative review, not a registered systematic review. The database backbone used dated native PubMed and Scopus searches, augmented by primary-source retrieval and targeted reference/citation searching. The gait PubMed query returned 752 records. Because connector pages overlapped, they contained only 428 distinct identifiers despite completion of all 16 pages. Official PubMed ESearch and EFetch reconciled all 752 records. Nine Scopus pages contained 210 unique records for a narrower title-focused query. These are search records, not counts of included studies; overlapping sources were not summed. Exact queries and limitations appear in the search appendix.

What the measures represent

Short walking capacity

A timed four- or ten-metre walk measures a specific episode of walking under an instruction and environment. Usual speed reflects a chosen pace; fastest-safe speed asks the person to mobilize reserve. Both depend on pain, confidence, effort, assistive devices and acceleration. A standing-start ten-metre test includes a different amount of acceleration from a middle-ten-metres test with five-metre run-ins. A four-metre timed section with two-metre run-ins is not a four-metre total course. Those distinctions can be large relative to the change a clinician wants to detect.

Kittelson and colleagues tested usual and fastest-safe four-metre walking on an eight-metre course, with timing over the central four metres. Two trials at each speed were recorded and average time used for reported performance. Their study provides a useful operational template, especially for early outpatient recovery, but the authors' preference for fastest speed is conditional on their population and purposes. It does not make usual speed irrelevant when the patient's habitual mobility or a usual-speed reference is the actual question. [3]

A 40-metre fast walk and a 50-foot walk are also legitimate tests, but are not interchangeable with a short straight walk. The older Kennedy protocol used two 20-metre lengths with the turn excluded from timing. Unver's 50-foot protocol used a single straight fast walk. A software interface should retain distance, start method, turn inclusion and instruction rather than collapsing all of these into a field called “gait speed.” [2, 5]

Sustained walking and composite mobility

The six-minute walk test (6MWT) quantifies distance accumulated under a sustained task. It combines locomotor capacity, pacing, symptoms and cardiorespiratory demand. Track configuration, turns, encouragement, rest and permitted aids are substantive parts of the measure. Kennedy used a 46-metre rectangular circuit; Kittelson used a 30.5-metre out-and-back walkway; Naylor used an outdoor 30-metre straight track. Distances from these protocols should not be treated as exchangeable without qualification. [2, 3, 8]

The two-minute walk test (2MWT) reduces testing burden, but shortening duration changes the construct and the opportunity for fatigue. Its instruction also varies: Sarac asked participants to walk as fast as possible on a 15-metre corridor, whereas Unnanuntana instructed normal-paced walking. A shorter test may be useful for someone unable to tolerate six minutes, but an improvement threshold from one version cannot simply be imported into the other. [4, 9]

TUG includes acceleration from standing, a three-metre walk, a turn, return and sitting. Chair characteristics, hand use and turn strategy affect its total time. In the literature, instructions range from normal or comfortable pace to fastest-safe pace. A slower TUG might reflect chair-rise difficulty or cautious turning rather than slow straight-line gait. Instrumented TUG can separate these components, but its component estimates require their own validation; a reliable total stopwatch time does not validate every algorithm-derived phase.

Walking quality and daily walking

Knee flexion excursion during loading response, knee extension moment, vertical ground-reaction force and spatiotemporal symmetry describe movement organization. They are not equivalent to walking speed. Interpretation requires the speed at which the measurements were obtained because faster walking changes mechanical demands. Comparisons at self-selected speed answer a different question from comparisons at a common imposed speed. The latter can reduce speed confounding while imposing a task that may not represent habitual walking.

A laboratory joint moment is usually estimated through inverse dynamics. It is not a direct measurement of quadriceps force, implant contact force or articular damage. A vertical-force ratio does not identify how the ankle, knee and hip share the demand. Similarly, a symmetrical walking pattern may arise because the contralateral limb has deteriorated. Absolute bilateral values and a healthy reference, when appropriate and protocol matched, should accompany ratios. Small biomechanical studies support the relevance of persistent unloading, but do not establish a universally harmful asymmetry boundary. [16, 18, 19]

Daily steps, walking-bout duration, cadence and sensor-derived speed describe real-world performance. They are affected by opportunity, weather, home layout, employment, goals and device wear, as well as physical capacity. A person who walks little may have adequate short-distance capacity; a person who accumulates many steps may do so slowly and with marked compensation. This is why capacity and activity should be displayed together, without suggesting that one validates the other. [17, 20, 21]

A practical selection framework

Table 1 A practical selection framework

The 2020 physical-therapy guideline recommends collecting KOOS JR with 30-second sit-to-stand and TUG at the first visit and at discharge in each care setting. Importantly, this outcome-assessment statement is designated best practice on insufficient evidence, not a high-certainty validation of every measure or threshold. Its motor-function recommendations support assessment and training of gait and movement symmetry, but do not supply a single optimal quantitative battery for every stage. [1]

Reliability and measurement error

Direct evidence for the four metre walk

The Kittelson analysis included 162 people awaiting primary unilateral TKA for OA, with assessments one to two weeks before surgery and at one, two, three, six and twelve months afterward. Participants were drawn from a rehabilitation trial. Important exclusions included BMI above 40, severe contralateral knee symptoms, function-limiting comorbidities and discharge to a facility other than home. Thus the sample does not represent every inpatient or frail TKA recipient. [3]

The reported fastest-walk ICCs ranged from 0.84 to 0.93, and the fastest-walk MDC95 ranged from 0.42 to 0.64 seconds across assessment occasions. At one month, the table reports ICC 0.93, SEM 0.23 seconds and MDC95 0.64 seconds; at one year, ICC 0.87, SEM 0.15 seconds and MDC95 0.42 seconds. These are time units, not metres per second. Usual-speed precision was generally poorer, with one-month MDC95 0.94 seconds. [3]

The critical design issue is that “test–retest” refers to two trials within the same testing session, separated by a short rest, usually under one minute. It is not a return visit on a different day with renewed swelling, analgesic timing, sleep, activity or confidence. In addition, the paper used ICC(2,1), while average trial time was the reported performance endpoint. The published values should be attributed to the authors' specific analysis, rather than silently described as reliability of any two-trial average.

Converting an error in time to an error in speed is nonlinear. At a fixed distance, speed equals distance divided by time. A decrease of 0.64 seconds has a different speed equivalent for someone taking three seconds than for someone taking eight seconds. Consequently, it is unsafe to convert the published fastest-walk MDC into one constant m/s cutoff for all patients.

Later bilateral TKA and the danger of stage transfer

Sarac and colleagues provide direct next-day reliability evidence, but in a narrow group: 41 patients in the seventh month after simultaneous primary bilateral TKA for severe OA, all able to stand and walk without auxiliary equipment. Testing was randomized in order, with practice and the same assessor on successive days. For the fast ten-metre walk, timed in the centre of a 20-metre path, Table 4 reports ICC 0.97, SEM 0.59 seconds and MDC95 1.63 seconds. For the fast 2MWT on a 15-metre corridor, the corresponding estimates are 0.94, 9.02 metres and 25 metres. [4]

These values are clinically informative for a matching population. They cannot be used as acute unilateral-TKA thresholds. The study excluded device-dependent patients; both knees had been replaced; and recovery was deliberately sampled after the most rapid early improvement. Its TUG table gives MDC95 3.04 seconds, whereas the discussion says 3.4 seconds. The discrepancy is disclosed rather than silently harmonized. More broadly, the study's correlations with the Berg Balance Scale and Falls Efficacy Scale establish shared variation among concurrent measures, not prediction of prospectively observed falls. [4]

Unver's lower-extremity inpatient reliability paper included THA, TKA, fracture and soft-tissue surgery. The abstract's overall results therefore must not be called TKA-only results. Its slow early postoperative performances illustrate why a time-based error band may be much larger than in mobile late-recovery samples. Where a TKA subgroup table is unavailable, it is better to label the mixed evidence and withhold a subgroup number than to reproduce a later paper's second-hand estimate. [22]

Same day walk and TUG studies

Yuksel and colleagues studied 48 TKA recipients using two trials separated by an hour. The original abstract reports ICC(2,1) 0.98 for TUG and 0.97 for 2MWT, with SEM/MDC95 of 0.82/2.27 seconds and 5.40/14.96 metres respectively. The full original was not retrieved for this review; precise stage, laterality distribution and operational details therefore remain restricted-access limitations. These abstract-verified values belong in an evidence table with that status, not in a universal postoperative interpretation rule. [6]

Jakobsen's 34-person 6MWT study similarly used two trials on one day with one hour of seated rest. The second trial was, on average, 14.1 metres longer. Its abstract reports ICC(2,1) 0.97, SEM 13.0 metres and smallest real difference 36.1 metres, and recommends the longer of two trials. The learning effect matters: a high ICC can coexist with systematic improvement on repeat exposure. Because the original full report was inaccessible, its exact recent-TKA interval and all course details have not been independently verified here. [7]

Unver's 2015 study has a more fundamental reporting problem. Thirty-three patients with bilateral TKA at least six months previously performed a fast straight 50-foot walk twice on the same day. The ICC was 0.97. Original Table 2, visually checked in the PDF, gives SEM 0.39 seconds and SRD95 0.91 seconds, whereas the abstract and results prose give SRD95 1.07 seconds. The authors' printed formula is SEM × 1.95 × √2. The table and prose disagreement should be preserved; recalculating a preferred value is not an authorized correction to the paper. The sample was relatively high-functioning and no participant used an aid, further limiting transport to early recovery. [5]

Why the older mixed arthroplasty benchmark remains useful

Kennedy's study is important because it examined both early deterioration and subsequent improvement, and documented feasibility. Its 150 participants included 81 knee and 69 hip replacements. Reliability, however, was estimated in only 17 stable preoperative patients, selected from a 21-person subset using stability on the Lower Extremity Functional Scale. Retest intervals were measured in months, not postoperative days. The frequently quoted MDC90 values are therefore mixed-joint preoperative estimates, not clean short-interval postoperative TKA error. [2]

The first postoperative assessment was within 15 days of surgery. The article reports a median eight-day interval from its preoperative to first postoperative assessment, and a median 38-day interval between postoperative assessments. Inability to complete a test is part of the clinical story, especially for longer walking and stairs. Analyses restricted to completers can make a test appear easier and can exclude those with the greatest early deficit. The report is most useful as a demonstration that reliability, responsiveness and feasibility require separate study designs. [2]

Measurement interpretation at a glance

Table 2 Measurement interpretation at a glance

Meaningful change and responsiveness

Detectable change and patient importance are separate

An SEM estimates measurement uncertainty in original units under a particular model. An individual MDC or SDC combines measurement error from repeated scores at a specified confidence level. A patient-important change asks whether the difference matters to the person or an appropriate clinical anchor. A statistical effect size expresses the size of a group change relative to variation. These are related concepts, but none can substitute automatically for another.

The practical problem is particularly acute after TKA because the patient is expected to change. Reliability estimated over a year of surgical recovery mixes stable rank ordering, heterogeneous recovery and real change. Conversely, excellent same-session reproducibility may understate the variability relevant to a clinic visit next week. A rehabilitation tool should show observed change first, then a matched error estimate when justified, and separately the patient's report of walking benefit or remaining limitation.

Six minute walking has relevant but imperfect anchor evidence

Naylor's analysis arose from the HIHO study, including randomized and observational rehabilitation groups. It included 243 primary unilateral OA-TKA recipients, with 158 available at ten weeks and 222 at 26 weeks. A seven-category patient transition question assessed walking change. The association between measured change and the anchor was modest: correlations for absolute distance were 0.297 from baseline to ten weeks and 0.259 from baseline to 26 weeks. There was essentially no association for the ten-to-26-week interval, so that interval did not support a threshold analysis. [8]

For slight improvement or more at 26 weeks, the absolute-distance threshold was 26 metres with Youden's method and 64.5 metres when targeting high specificity. The ROC area was 0.724. At 26 metres, sensitivity was approximately 73% and specificity 65%; at 64.5 metres, sensitivity fell to 50% while specificity was approximately 83%. The difference between the two thresholds is not an inconsistency: they answer different classification priorities. [8]

Nevertheless, the anchor groups were highly imbalanced, the authors had to combine slight with larger improvement, and the anchor-change correlations were weaker than the commonly desired level. The resulting range is useful as a context-specific interpretation aid for preoperative-to-six-month improvement; it is not a precise individual minimum, a deterioration threshold, or evidence for a later six-week rehabilitation episode. The study itself demonstrates why one unqualified “6MWT MCID” is inadequate.

A distribution derived two minute walk threshold

Unnanuntana's prospective cohort enrolled 162 primary TKA patients and retained 157 at one year, including 24 simultaneous bilateral procedures. The study excluded complex procedures and immediate postoperative complications, which limits representation of difficult early recovery. The 2MWT used normal-paced walking with aids permitted. Mean distance increased from 46.2 metres before surgery to 66.1 metres at twelve months. Large group effect size and standardized response mean support responsiveness in this sample. [9]

The reported 12.7-metre “MCID, ” however, was calculated as baseline SD × √(1−r), with r defined as the ICC between baseline and one-year postoperative performance. The paper used the same approach to derive approximately 9.5 seconds for TUG. This is a distribution-derived calculation without a patient-importance anchor; using a recovery interval for the ICC also prevents it from being interpreted as ordinary stable-state measurement error. The study provides valuable longitudinal data, but its label does not establish an individual patient-important threshold. This distinction should remain explicit wherever the number is cited. [9]

New digital gait estimates need further verification

Redfern and colleagues' 2026 secondary analysis is directly relevant to digital rehabilitation. It examined smartphone/watch-derived metrics after primary elective OA-TKA. The source cohort included 3,403 TKA recipients at 28 sites; the gait-speed anchor analysis involved 684 with both gait and EQ-5D-5L data, and a larger gait-data cohort numbered 971. Simultaneous or closely staged bilateral procedures were excluded. Users had to own a compatible iPhone, and the study supplied a watch. [10]

The retrieved body reports an anchor-based gait-speed MCID of 0.067 m/s at one year, using EQ-5D-5L change categories derived from distributional effect-size conventions. This is not a direct global rating that a small change in walking was important. It is also a HealthKit-derived free-living measure, not a stopwatch clinic walk. Baseline function, data completeness, technology selection and the distinction between quality-of-life and walking anchors limit portability. The authors acknowledge that several proposed MCIDs were below estimated MDCs. [10]

Most importantly, the retrieved body states that 281 of 971 patients, or 28.9%, achieved the gait-speed MCID, while the indexed abstract reports 79%. Units for percentage-based gait metrics also differ between abstract and body expressions. The original numerical tables/figures were not recovered sufficiently to settle these conflicts. The report therefore treats the 0.067-m/s estimate as preliminary and explicitly withholds the disputed attainment rate and percentage thresholds from deployment. This is a source-verification issue, not a reason to ignore digital outcomes or to substitute thresholds from stroke, OA or THA.

Recovery does not imply normalization

Kittelson's fastest short walk and 6MWT were responsive to the first two months of recovery. After three months, short-walk responsiveness was small, while the longer walk still detected later gains. This supports matching the test to the stage and recovery question. It does not show that every patient has plateaued at three months or recovered a healthy gait pattern. [3]

The Tolk cohort likewise separates different measurement questions: preoperative reliability was assessed in a 30-person subgroup after only 30 minutes, whereas responsiveness covered surgery and twelve months of recovery. Its 40-metre walk fulfilled more responsiveness hypotheses than its chair stand, despite contested baseline construct validity. A test can be reproducible without strongly matching a questionnaire, and can be responsive over a major surgical interval without being sensitive to a small treatment change. [23]

Prognosis and later outcomes

Preoperative prognosis and early postoperative updating

Preoperative function is a useful baseline, but it is not a ceiling on recovery or an argument to deny surgery. It can relate more strongly to later absolute status than to change, because someone starting with better capacity has less numerical room to improve. Models of change require particular care with mathematical coupling and regression to the mean.

Mizner's original 40-person preoperative study followed patients after primary unilateral TKA and linked preoperative quadriceps strength to performance-based function one year later. This supports evaluating muscle capacity alongside mobility. It does not imply that increasing a measured strength value by a given amount will produce the regression-predicted improvement in walking, nor that strength alone forecasts patient-reported benefit. The small development sample and absence of independent validation constrain individual use. [15]

Zeni and Snyder-Mackler followed 155 people from an early outpatient assessment to one and two years. Initial mobility and contralateral quadriceps strength contributed to later function, emphasizing that the nonoperated leg is not merely a denominator or control. This is genuine temporal evidence, but selected unilateral cohorts with limited contralateral symptoms do not capture all bilateral disease presentations. Their models also predict functional tests and a questionnaire, not implant survival or every aspect of community participation. [14]

Pua's 2016 study is a stronger multivariable example. Among 1,096 patients with six-month outcomes, approximately 12% reported a maximum walking time of fifteen minutes or less. Predictors included preoperative walking limitation, BMI, contralateral knee pain, preoperative aid use and fast ten-metre speed measured four weeks after surgery. The model's optimism-corrected concordance was 0.71, with bootstrap internal validation and calibration assessment. [11]

The timing matters: this is not a wholly preoperative model because it needs a one-month gait result. Thirty per cent lacked the four-week speed/pain measurement, addressed by multiple imputation. The outcome was self-reported walking-time limitation, not a measured endurance distance. The authors explicitly called for external validation. Its clinical lesson is that early measured mobility can update prognosis; its published equation should not be silently transported into a different service or a different test instruction. [11]

Pua's later 4,026-person cohort developed preoperative models for six-month range of motion, pain and walking limitation. These outcomes were modelled separately. The cohort's dates and centre overlap with the earlier work, so it is not an independent replication of the 2016 population. It expands the prognostic evidence but should not be counted twice as independent confirmation. [12]

Temporally tested recovery trajectories

Kim and colleagues developed a neighbours-based TUG prediction approach in 397 people and tested it in a temporally distinct group of 202. Matching used age, sex, BMI and preoperative TUG, with 35 comparable patients selected in development. The model represented recovery over the first six months and reported calibration in the later dataset. It is a more persuasive prediction design than an unvalidated regression fitted and assessed in the same people. [13]

The median prediction and its uncertainty need to travel together. The often-quoted 2.33-second width was the average 50% prediction interval in model tuning, not a 95% confidence band and not a measurement-error threshold. Half of observations are expected outside a correctly calibrated central 50% prediction interval. The later test sample came from the same small collection of clinical and research sources, so temporal validation is not proof of geographic or health-system transport. The source data required both preoperative and postoperative TUG and had substantial missingness. A model should be tested against local care, population and protocol before clinical implementation. [13]

The subsequent comparison of “people-like-me” and linear mixed models is relevant methodological work, and its original article body has now been recovered. The completed audit checked the comparison, the 317 training and 456 testing records, postoperative inputs and geographic/methodological test separation; the supplement was not separately inspected. These checks do not establish broad clinical benefit. [24]

Future gait mechanics are not the same as future walking capacity

The 2025 gait-biomechanics analysis included 126 participants in a movement-training trial. It modelled six-month surgical-limb knee extension moment, knee flexion excursion and force ratio from preoperative or ten-week variables. Baseline mechanics and quadriceps-related variables mattered, but which factors entered the model depended on the mechanical outcome. A clinical quadriceps activation score collected at the postoperative treatment evaluation was included even in the model labelled “preoperative.” It is therefore important to inspect predictor availability before using that model for preoperative counselling. [16]

The analyses were exploratory, with univariate screening and backward selection, and did not establish an externally validated bedside calculator. Eligibility also excluded substantial contralateral disease, prior contralateral TKA, function-limiting comorbidity and inability to walk thirty metres safely without an aid. Treatment assignment influenced force-ratio results and is part of the context. A model derived inside a rehabilitation trial cannot automatically describe recovery under all pathways. [16]

Kline's earlier 24-person study measured rate of torque development and gait at three and six months. Different same-visit regressions linked quadriceps RTD to knee excursion at each occasion. Although the cohort was followed over time, these regressions did not test whether three-month RTD predicted six-month gait. Sixteen candidate predictors in a small sample, stepwise selection and 100-Hz dynamometer torque acquisition (gait ground-reaction forces were sampled separately at 1200 Hz) further limit precision and mechanistic inference. The study motivates hypotheses about rapid force production; it does not supply a validated RTD cutoff for gait recovery. [18]

Later physical activity

A 2025 secondary analysis of a Veterans rehabilitation trial is useful because the outcome occurred later and outside the clinic. Capacity was tested at fourteen weeks, and thigh-accelerometer daily stepping and peak cadence were assessed at 38 weeks. The report included 87 predominantly male participants, with smaller denominators in fitted models because of missing predictors. TUG and a physical-health questionnaire were associated with later daily steps; TUG and assignment to a physical-activity intervention were associated with later peak cadence. The corresponding models explained only a modest portion of variation. [17]

The 30-second chair stand was not associated with these later activity measures in the preliminary regressions, and no significant predictor of achieving the selected 7,500-steps/day category was identified. This is a useful negative result: a valid clinic performance measure does not necessarily forecast how active a patient will become. Rehabilitation exposure, behaviour and environment remain important. A model developed to predict a continuous step count cannot be assumed to classify a health target accurately. [17]

Prognostic evidence at a glance

Table 3 Prognostic evidence at a glance

Instrumented and digital assessment

Agreement has to be established for each variable

A sensor's validity is not an all-or-nothing property. Bravi's study of twenty recent THA/TKA patients walking with crutches and ten healthy comparators found more encouraging results for general spatiotemporal parameters than for gait-cycle phases. The abstract reports particularly weak patient correlations for some support-phase measures. Because the original full text was not retrieved, these findings are used to identify the limitation rather than to authorize a numerical device accuracy claim. Mixed joints, crutches and intertrial reliability must remain visible. [25]

Correlation with a reference does not establish agreement. Two systems can rank patients similarly while systematically disagreeing about speed, stance time or asymmetry. For a proposed clinic device, inspect bias, limits of agreement, error across slow and fast walking, missing-event rates, reference synchronization and the exact output definition. The validity of cadence does not validate knee angles; the validity of angles does not validate estimated joint moments.

Walking aids can change what a wrist device records

The 2024 consumer-monitor study tested twenty-three primary OA-TKA recipients on a fifty-metre hallway, before surgery without an aid, at postoperative days one to three with a walker, at six weeks with a cane and at six months without an aid. Fitbit Charge and Apple Watch Series 4 step counts were compared with an observer's tally. Performance was much poorer with a walker, and some devices recorded zero steps. This is directly relevant to early remote monitoring, where a zero may mean an algorithm failure rather than inactivity. [20]

The study is small and concerns specific hardware/software generations. Aid condition and postoperative stage changed together, so it does not isolate their causal effects. Nonetheless, it establishes that acceptable no-aid accuracy cannot be assumed in walker-dependent recovery. Device name, placement, aid use, wear time and algorithm version should accompany longitudinal trends. A step increase after stopping use of a walker may partly reflect improved capture.

Continuous acceleration and emerging models

Ghaffari's thirty-one-person study recorded thigh accelerometry from two weeks before unilateral TKA to ninety days afterward. Temporal and frequency-domain features changed across recovery, and some related to health-status improvement. This shows feasibility and potentially useful responsiveness, but not a validated prediction of future disability. The cohort excluded frailty, walking-aid users and severe contralateral disease. The classifier work was small, with repeated measurements and model-fitting limitations; a favourable apparent ROC area should not become a clinical triage cutoff. [21]

For markerless video, applicability must be demonstrated in postoperative patients under the intended view, clothing, aid and speed conditions. The search identified promising adjacent work, including knee-OA angle validation and healthy-participant testing of systems intended for TKA rehabilitation. Neither supplies postoperative TKA validity by itself. A practical video tool can document observable features and timing, while labelling unvalidated force or loading estimates as exploratory.

Stage specific interpretation and implementation

During inpatient recovery, the first question is whether the person can mobilize safely with the assistance available. Record bed/chair transitions, walking distance achieved, aid and person assistance, and the reason a standardized test was not completed. A noncompleted 6MWT is not six minutes at zero speed, and an assisted TUG is not a standard independent TUG. Changing analgesia, swelling, orthostatic symptoms and fatigue can dominate very short-term change.

During early outpatient recovery, choose a feasible short-walk protocol and keep it stable. If the aid changes, retain that change in the record and display the scores as different conditions. A faster result with a walker can represent useful supported capacity; an unaided result may answer a different question. Add TUG or an endurance measure according to the specific limitation rather than simply increasing the test count.

At three to six months, compare walking capacity with mechanics and real-world goals. A short-walk plateau alongside improvement in six-minute distance need not be contradictory. Better speed with limited loading-response knee flexion warrants a movement-quality assessment, but the size of asymmetry alone cannot identify a surgical problem or prescribe an intervention. At twelve months and beyond, reassess contralateral OA, comorbidity, activity opportunities and patient priorities; do not define recovery solely as return to the already-impaired preoperative state.

A defensible assessment record contains the surgery date and indication; laterality and contralateral status; test version, pace and course; timing landmarks and equipment; number of practice and scored trials; trial aggregation; footwear and aids; person assistance; symptoms and medication timing where relevant; completion or stopping status; and the exact units. Instrumented outputs additionally need device and software version, placement, sampling, processing, quality checks and the anatomical/event definitions.

The interface should show raw scores and conditions before interpretation. It may display “within the range observed in this study” with a clearly matched reference, but should not equate a percentile with normality, a prediction interval with an error band, or an MDC exceedance with important benefit. Automated flags should remain conservative when the reference population is bilateral, mixed-joint, abstract-only, or subject to unresolved source inconsistency.

Evidence limitations and priorities

The strongest practical evidence supports measuring walking in a standardized way and using multiple complementary perspectives. The main gaps are between-day error during rapid early recovery; validated patient-important change under exact clinic and free-living protocols; inclusion of frail and assisted patients; external validation of prognostic models; and whether use of such models improves decisions or outcomes. Most biomechanical cohorts are small and selected. Several prominent datasets recur across publications, limiting the number of independent replications.

Future work should prespecify the intended outcome, measurement stage and decision. It should retain unsuccessful tests, explain missing device data, compare algorithms with appropriate reference standards and validate models on genuinely new patients. Calibration, decision utility and uncertainty matter as much as discrimination. Rehabilitation trials can show treatment effects, but they do not alone establish measurement error, prediction accuracy or an optimal personal target.

Conclusion

After primary OA-TKA, a standardized short walk is an efficient starting point, while TUG, sustained walking, movement analysis and free-living monitoring answer complementary questions. Interpretation depends on protocol, postoperative stage, assistance and the contralateral limb. Direct TKA evidence is preferable to imported OA, THA or neurological thresholds, but even TKA-labelled evidence may be bilateral, mixed-stage or methodologically unsuitable for a particular patient. The safest clinical tool preserves those distinctions and communicates change and prognosis with their uncertainty.

Primary study evidence matrix

This matrix is a compact audit companion, not a pooled estimate or numerical quality score. A denotes measurement properties; B concurrent association; C longitudinal recovery or association; D prediction development; E prediction with separate testing. Values retain original units and confidence levels.

Table 4 Primary study evidence matrix

[1] is contextual clinical consensus rather than primary validation. [24] is a prediction-method extension whose original body has now been recovered and scoped comparison/validation details checked; its supplement was not separately inspected. No pooled result is calculated, and overlapping cohorts are not treated as independent validation.

Search appendix

Exact executed PubMed query

("Arthroplasty, Replacement, Knee"[MeSH Terms] OR "total knee arthroplasty"[Title/Abstract] OR "total knee replacement"[Title/Abstract]) AND (gait[Title/Abstract] OR walk*[Title/Abstract] OR "timed up and go"[Title/Abstract]) AND (reliab*[Title/Abstract] OR valid*[Title/Abstract] OR responsiv*[Title/Abstract] OR "measurement error"[Title/Abstract] OR "minimal detectable"[Title/Abstract] OR "minimal important"[Title/Abstract] OR prognos*[Title/Abstract] OR predict*[Title/Abstract] OR longitudinal[Title/Abstract]) AND ("1800/01/01"[Date - Publication] : "2026/10/02"[Date - Publication])

Exact executed Scopus query

TITLE("knee arthroplasty" OR "knee replacement") AND TITLE(gait OR walk* OR "timed up and go" OR sensor* OR wearable* OR markerless) AND TITLE-ABS-KEY(reliab* OR valid* OR responsiv* OR "measurement error" OR "minimal detectable" OR "minimal important" OR prognos* OR predict* OR longitudinal) AND PUBYEAR < 2027

Coverage and reconciliation

Scopus reported 210 records. The returned pages contained 210 distinct Scopus identifiers. This query is title-focused and narrower than PubMed, not a second identical search. PUBYEAR < 2027 is a yearly restriction; electronically published papers were checked against the actual cutoff where relevant.

Counts are retrieval records, not included studies. Sources and domains overlap and cannot be summed. Deduplication used DOI, PMID/provider ID and title/author/year. Treatment-only, technical implant wear, off-indication, revision and mixed-population records were screened for report relevance rather than removed by aggressive query exclusions.

Targeted follow on work

Reference and citation discovery was used for Kennedy 2005, Christiansen 2011 and Pua 2016; these provider-limited calls are discovery, not exhaustive citation coverage. Original retrieval targeted each cited DOI through bibliographic databases, official PMC and publisher pages, and author or institutional repositories. A targeted search located the Huber thesis chapter, Farquhar correction and recent Blessinger companion papers/correction. Failed or restricted retrieval stayed explicitly marked. No access was inferred from a metadata open-access flag.

Main limitations

The measurement/prognosis terminology and title-focused Scopus strategy can miss biomechanical or technology studies not using these labels. Targeted chasing reduces but does not eliminate this problem. The manuscripts therefore characterize the evidence appraised and its limits rather than claim exhaustive absence of studies. Later publications are not included.

References

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

1. Jette, Diane U, Hunter, Stephen J, Burkett, Lynn, Langham, Bud, Logerstedt, David S, Piuzzi, Nicolas S, et al. Physical Therapist Management of Total Knee Arthroplasty. Physical therapy. 2020. DOI 10.1093/ptj/pzaa099 Source examined: Complete article; consensus statement, not primary validation.

Source note: SRC-27ba0555a48c Jette 2020

2. Kennedy DM, Stratford PW, Wessel J, Gollish JD, Penney D. Assessing stability and change of four performance measures: a longitudinal study evaluating outcome following total hip and knee arthroplasty. BMC musculoskeletal disorders. 2005;6:3. DOI 10.1186/1471-2474-6-3 Source examined: Complete original article and tables.

Source note: SRC-002a53da5f5c Kennedy DM 2005

3. Kittelson A, Carmichael J, Stevens-Lapsley J, Bade M. Psychometric properties of the 4-meter walk test after total knee arthroplasty. Disability and rehabilitation. 2022;44(13):3204-3210. DOI 10.1080/09638288.2020.1852446 Source examined: Complete original article and Table 2.

Source note: SRC-c2e936ed5bac Kittelson A 2022

4. Sarac DC, Unver B, Karatosun V. Validity and reliability of performance tests as balance measures in patients with total knee arthroplasty. Knee surgery & related research. 2022;34(1):11. DOI 10.1186/s43019-022-00136-4 Source examined: Complete original article and tables; TUG table/discussion discrepancy retained.

Source note: SRC-7757dfb24f28 Sarac DC 2022

5. Unver B, Kalkan S, Yuksel E, Kahraman T, Karatosun V. Reliability of the 50-foot walk test and 30-sec chair stand test in total knee arthroplasty. Acta ortopedica brasileira. 2015;23(4):184-7. DOI 10.1590/1413-78522015230401018 Source examined: Original article; Table 2 visually checked.

Source note: SRC-5add70458765 Unver B 2015

6. Yuksel E, Kalkan S, Cekmece S, Unver B, Karatosun V. Assessing Minimal Detectable Changes and Test-Retest Reliability of the Timed Up and Go Test and the 2-Minute Walk Test in Patients With Total Knee Arthroplasty. The Journal of arthroplasty. 2017;32(2):426-430. DOI 10.1016/j.arth.2016.07.031 Source examined: Abstract verified; full article unavailable.

Source note: SRC-a5c17bfae5d5 Yuksel E 2017

7. Jakobsen TL, Kehlet H, Bandholm T. Reliability of the 6-min walk test after total knee arthroplasty. Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA. 2013;21(11):2625-8. DOI 10.1007/s00167-012-2054-y Source examined: Abstract verified; full article unavailable.

Source note: SRC-b6fdb3fbdb86 Jakobsen TL 2013

8. Naylor JM, Mills K, Buhagiar M, Fortunato R, Wright R. Minimal important improvement thresholds for the six-minute walk test in a knee arthroplasty cohort: triangulation of anchor- and distribution-based methods. BMC musculoskeletal disorders. 2016;17(1):390. DOI 10.1186/s12891-016-1249-7 Source examined: Complete original article and tables.

Source note: SRC-92965a184951 Naylor JM 2016

9. Unnanuntana A, Ruangsomboon P, Keesukpunt W. Validity and Responsiveness of the Two-Minute Walk Test for Measuring Functional Recovery After Total Knee Arthroplasty. The Journal of arthroplasty. 2018;33(6):1737-1744. DOI 10.1016/j.arth.2018.01.015 Source examined: Complete original article; method and result sections directly appraised.

Source note: SRC-6591c209e1d1 Unnanuntana A 2018

10. Redfern RE, Khan ST, Archibeck MJ, Peters CL, Anderson MB, Piuzzi NS. Establishing Minimal Clinically Important Difference and Substantial Clinical Benefit Thresholds for Objective Gait Metrics After Total Knee Arthroplasty. The Journal of arthroplasty. 2026;41(10S1):S191-S197. DOI 10.1016/j.arth.2026.03.101 Source examined: Complete retrieved body but original numerical tables/figures unavailable; body/abstract discrepancy unresolved.

Source note: SRC-17280cbb2ded Redfern RE 2026

11. Pua YH, Seah FJ, Clark RA, Poon CL, Tan JW, Chong HC. Development of a Prediction Model to Estimate the Risk of Walking Limitations in Patients with Total Knee Arthroplasty. The Journal of rheumatology. 2016;43(2):419-26. DOI 10.3899/jrheum.150724 Source examined: Complete original article.

Source note: SRC-b5012dcd12e3 Pua YH 2016

12. Pua YH, Poon CL, Seah FJ, Thumboo J, Clark RA, Tan MH, et al. Predicting individual knee range of motion, knee pain, and walking limitation outcomes following total knee arthroplasty. Acta orthopaedica. 2019;90(2):179-186. DOI 10.1080/17453674.2018.1560647 Source examined: Complete original article; supplemental material not independently reconstructed.

Source note: SRC-4f5122e87122 Pua YH 2019

13. Kim C, Colborn KL, van Buuren S, Loar T, Stevens-Lapsley JE, Kittelson AJ. Neighbors-based prediction of physical function after total knee arthroplasty. Scientific reports. 2021;11(1):16719. DOI 10.1038/s41598-021-94838-6 Source examined: Complete original article; model supplement not independently reproduced.

Source note: SRC-1a40cef114b2 Kim C 2021

14. Zeni JA Jr, Snyder-Mackler L. Early postoperative measures predict 1- and 2-year outcomes after unilateral total knee arthroplasty: importance of contralateral limb strength. Physical therapy. 2010;90(1):43-54. DOI 10.2522/ptj.20090089 Source examined: Complete original article.

Source note: SRC-95a9e043b7c1 Zeni JA Jr 2010

15. Mizner RL, Petterson SC, Stevens JE, Axe MJ, Snyder-Mackler L. Preoperative quadriceps strength predicts functional ability one year after total knee arthroplasty. The Journal of rheumatology. 2005;32(8):1533-9. PubMed. Source examined: Complete original article.

Source note: SRC-c1003e7fcd2a Mizner RL 2005

16. Capin JJ, Zeni JA Jr, Forster JE, Cheuy VA, Peters A, Hogan C, et al. Preoperative and Post-Rehabilitation Predictors of Gait Biomechanics Six Months After Total Knee Arthroplasty. Journal of orthopaedic research : official publication of the Orthopaedic Research Society. 2025;43(11):1964-1972. DOI 10.1002/jor.70052 Source examined: Complete original article; supplement not independently reanalysed.

Source note: SRC-24097ae72925 Capin JJ 2025

17. Kline PW, Hanlon SL, Richardson VL, Hoffman RM, Melanson EL, Juarez-Colunga E, et al. Functional Capacity at Rehabilitation Discharge Predicts Physical Activity Characteristics 24 Weeks Later for People With Total Knee Arthroplasty: A Secondary Analysis of a Randomized Controlled Trial. Archives of physical medicine and rehabilitation. 2025;106(6):845-852. DOI 10.1016/j.apmr.2025.01.416 Source examined: Complete original article; later-outcome analysis directly appraised.

Source note: SRC-c0b74bd10b4b Kline PW 2025

18. Kline PW, Jacobs CA, Duncan ST, Noehren B. Rate of torque development is the primary contributor to quadriceps avoidance gait following total knee arthroplasty. Gait & posture. 2019;68:397-402. DOI 10.1016/j.gaitpost.2018.12.019 Source examined: Complete original article and tables.

Source note: SRC-b2e50f266f12 Kline PW 2019

19. Mizner RL, Snyder-Mackler L. Altered loading during walking and sit-to-stand is affected by quadriceps weakness after total knee arthroplasty. Journal of orthopaedic research : official publication of the Orthopaedic Research Society. 2005;23(5):1083-90. DOI 10.1016/j.orthres.2005.01.021 Source examined: Abstract verified; full article unavailable.

Source note: SRC-98f4579514d6 Mizner RL 2005

20. Kooner P, Baskaran S, Gibbs V, Wein S, Dimentberg R, Albers A. Commercially available activity monitors such as the fitbit charge and apple watch show poor validity in patients with gait aids after total knee arthroplasty. Journal of orthopaedic surgery and research. 2024;19(1):404. DOI 10.1186/s13018-024-04892-9 Source examined: Complete original article; specific commercial devices.

Source note: SRC-11f1062f1771 Kooner P 2024

21. Ghaffari A, Clasen PD, Kappel A, Rasmussen J, Gurchiek RD, Kold S, et al. Monitoring Gait Recovery After Total Knee Arthroplasty Using Wearable Sensors: Responsiveness of Gait Accelerations. Journal of orthopaedic research : official publication of the Orthopaedic Research Society. 2025;43(12):2165-2177. DOI 10.1002/jor.70058 Source examined: Complete original article; exploratory wearable study.

Source note: SRC-327bc0c030e9 Ghaffari A 2025

22. Unver B, Baris RH, Yuksel E, Cekmece S, Kalkan S, Karatosun V. Reliability of 4-meter and 10-meter walk tests after lower extremity surgery. Disability and rehabilitation. 2017;39(25):2572-2576. DOI 10.1080/09638288.2016.1236153 Source examined: Abstract verified; mixed lower-extremity surgery, original subgroup table not retrieved.

Source note: SRC-0985a6d897e9 Unver B 2017

23. Tolk JJ, Janssen RPA, Prinsen CAC, Latijnhouwers DAJM, van der Steen MC, Bierma-Zeinstra SMA, et al. The OARSI core set of performance-based measures for knee osteoarthritis is reliable but not valid and responsive. Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA. 2019;27(9):2898-2909. DOI 10.1007/s00167-017-4789-y Source examined: Original article and Table 2.

Source note: SRC-a6727b5507a5 Tolk JJ 2019

24. Graber J, Kittelson A, Juarez-Colunga E, Jin X, Bade M, Stevens-Lapsley J. Comparing "people-like-me" and linear mixed model predictions of functional recovery following knee arthroplasty. Journal of the American Medical Informatics Association : JAMIA. 2022;29(11):1899-1907. DOI 10.1093/jamia/ocac123 Source examined: Original article body now recovered; scoped comparison, input and testing-separation details checked. Supplement not separately inspected.

Source note: SRC-2312f20a93d0 Graber J 2022

25. Bravi M, Gallotta E, Morrone M, Maselli M, Santacaterina F, Toglia R, et al. Concurrent validity and inter trial reliability of a single inertial measurement unit for spatial-temporal gait parameter analysis in patients with recent total hip or total knee arthroplasty. Gait & posture. 2020;76:175-181. DOI 10.1016/j.gaitpost.2019.12.014 Source examined: Abstract verified; full article unavailable.

Source note: SRC-7793ba556d88 Bravi M 2020