Does Multimodal Rehabilitation for Ankle Instability Improve Patients’ Self-assessed Functional Outcomes? A Network Meta-analysis

Abstract

Background

Although there are many nonsurgical treatment options for the primary management of chronic ankle instability, the most effective nonoperative intervention has not been defined.

Questions/purposes

The purpose of this study was to perform a network meta-analysis to compare the results of different standalone and/or combined nonsurgical interventions on chronic ankle instability as measured by (1) the Cumberland Ankle Instability Tool (CAIT) at 0 to 6 months after treatment and (2) treatment-related complications.

Methods

We searched PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), and Scopus in August 2017 for completed studies published between 2005 and 2016. We conducted random-effects pairwise and network meta-analysis considering randomized trials, which compared the effects of various nonoperative therapies for ankle instability. Studies assessing patients with functional ankle instability and/or mechanical ankle instability and/or recurrent ankle sprains were eligible for inclusion. After combining data from self-administered questionnaires, we analyzed patient self-reported outcomes of function at the end of the rehabilitation period and 1 to 6 months after treatment. We thereafter reexpressed standardized mean differences to mean differences with CAIT. For this instrument, scores vary between 0 and 30, and higher scores indicate better ankle stability. We included 21 trials involving 789 chronically unstable ankles. The rehabilitation interventions included, but were not limited to, balance training, strengthening exercises, a combination of the balance and strengthening exercises, manual therapy, and multimodal treatment. The implemented multistation protocols were targeted at four main areas of rehabilitation (ROM, balance, strength, and overall activity). Control was defined as placebo and/or wait and see. Treatment-related complications were defined as any major or minor adverse event observed after rehabilitation as reported by the source studies. Statistically, we did not detect significant inconsistency in the network meta-analysis. We also assessed the quality of the trials using the Cochrane risk of bias tool and judged 12, eight, and one studies to be at a low, unclear, and high risk of bias, respectively. We also considered the quality of evidence to be of moderate strength utilizing the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) approach. We defined the minimum clinically important difference (MCID) in the CAIT to be 3 points.

Results

A 4-week supervised rehabilitation program, which included balance training, strengthening, functional tasks, and ROM exercises, was favored over control according to the results of four trials by a clinically important margin (mean difference between multimodal and control groups in the CAIT was -10; 95% confidence interval [CI], -16 to -3; p = 0.001). Among the standalone interventions, only balance training was better than control according to the findings of seven trials (mean difference between balance training and control in the CAIT was -5; 95% CI, -10 to -0.03; p = 0.049); this difference likewise exceeded the MCID and so is believed to be a clinically important difference. Adverse events associated with the enrolled rehabilitation protocols were transient, mild, and uncommon.

Conclusions

Although a supervised impairment-based program after chronic ankle instability was superior to control, we note that followup in the included trials tended to be short and inconsistent, although the effect size exceeded the MCID and so likely would be identified as clinically important by patients. Future randomized trials should determine whether the short-term benefits of these interventions are sustained over time.

Level of Evidence

Level I, therapeutic study.

Introduction

Studies suggest that ankle sprains of the lateral complex are the most common lower extremity injury in physically active persons [30, 42]. Approximately two million ankle sprains occur every year in the United States, which results in USD 2 billion in healthcare costs [54]. Many patients with lateral ankle sprains have ongoing symptoms, including pain, giving-way episodes, and instability [56], which can be disabling. These symptoms represent the principal features of chronic ankle instability [22]. An estimated 32% to 74% of patients with lateral ankle sprains develop instability [2, 34].

Patients with ankle instability should initially be treated nonoperatively [1]. However, there is no consensus regarding the most effective nonoperative treatment of ankle instability. We therefore performed a network meta-analysis (NMA) to assess patient-reported evidence, which is recognized as a critical component of evidence-based medicine [14, 53]. An NMA study design allows us to rank the competing treatments evaluated by prior randomized trials, even when those individual treatments were not compared against one another in the source studies. By doing so, we hoped to build on the results of previous meta-analyses [35, 45], which demonstrated that balance training provided consistent self-reported improvements in patients with ankle instability. The authors of these meta-analyses also highlighted the need for the identification of the most effective rehabilitation intervention for optimizing self-based outcomes of function in individuals with ankle instability.

More specifically, we sought to perform a NMA to compare the different standalone and/or combined rehabilitation interventions for chronic ankle instability as measured by (1) the Cumberland Ankle Instability Tool (CAIT) at 0 to 6 months after treatment and (2) treatment-related complications.

Patients and Methods

We prospectively registered this systematic review with PROSPERO (CRD42016037849) and report the results of an included prespecified outcome analysis. We used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for NMA [31]. Our institution waived approval for the reporting of this investigation because this study is a systematic review of the literature.

Inclusion Criteria

In this NMA, we developed our inclusion criteria based on the Hiller et al. [28] model, which consists of nine subgroups of chronic ankle instability and three deficit classes. Accordingly, studies assessing patients with functional ankle instability and/or mechanical ankle instability and/or recurrent ankle sprains were eligible for inclusion. For each of these subgroups, we accepted the diagnosis of lateral ankle instability provided by the authors of the included studies. We considered trials that used at least one self-administered questionnaire as an outcome measurement that evaluated instability, function, or disability. Concerning the eligible study designs, we enrolled multiarm, parallel-group, and nested randomized trials comparing the effects of at least one nonsurgical intervention for unilateral ankle instability.

Exclusion Criteria

We excluded trials that compared the results of operative interventions for ankle instability because the indications for surgical management are different [1]. To elaborate, the rationale behind nonoperative management of ankle instability is to prevent the need for surgical stabilization. In addition, we did not consider studies that explored the effects of passive restraints (ie, taping and/or bracing) or trials on acute ankle sprains. A passive restraint changes the network of forces in and around the ankle during normal motion and physical exertion and therefore would change the effect of the intervention.

Information Sources and Search

Using no language restrictions, two review authors (KT, DG) performed an electronic database and manual search in a blinded fashion to identify published and unpublished completed randomized trials on August 26, 2017; we sought to identify completed studies published between 2005 and 2016. In particular, we searched PubMed (Table 1), Scopus, and the Cochrane Central Register of Controlled Trials (CENTRAL). We also considered the following trial registries: International Standard Randomized Controlled Trial Number (ISRCTN) Register, ClinicalTrials.gov, and Australian New Zealand Clinical Trials Registry (ANZCTR). Finally, we examined reference lists of relevant systematic reviews and conference abstracts.

Table 1.

Search strategy for PubMed using Boolean operators

Our systematic search revealed 668 potentially relevant records. After deleting duplicates, we screened the remaining 523 records for inclusion. Based on information provided in the abstract and the title, 30 articles met our eligibility criteria. One published record [60] and four trial registries (NCT00601471, NCT01298856, NCT01541657, NCT01790581) referred to duplicates of published randomized controlled trials. We also found three eligible trial protocols (ACTRN12616000234415, ACTRN12616000386437, NCT02945943) without available results. One study referred to a nonrandomized trial [23]. The results of the study selection procedure are presented in the flowchart of the present systematic review (Fig. 1). One conference abstract was eligible for inclusion in this NMA [32].

Fig. 1.

The flowchart of the study selection procedure is presented.

Study Characteristics

Two investigators (KT, HSV) independently abstracted information concerning the comparators in the intervention groups, countries in which the studies took place, anthropometrics and activity level of the enrolled patients, outcomes, and followup measurements. Data extraction also included information about potential adverse events, inclusion criteria, supervision of the participants, and chronicity of the disease. If information was missing, we contacted the corresponding authors of the included trials to request their data. Two review authors (KT, DM) resolved any discrepancies.

In the current systematic review, we included 21 studies with a total of 789 participants (Table 2). Of these, 11 trials were conducted in the United States [4, 5, 10, 17, 25, 27, 32, 40, 41, 51, 61], five in Europe [8, 11, 12, 44, 50], four in Asia [3, 16, 33, 36], and one in Africa [37]. Except for one nested randomized trial [25], all the included studies utilized a standard study design. For the included studies, the reported mean age and height of the participants ranged from 18 to 34 years and between 166 cm to 178 cm, respectively. The mean reported weight varied between 65 and 82 kg (see Table, Supplemental Digital Content 1).

Table 2.

Eligibility criteria and intervention-associated characteristics of the included trials

Outcome Assessment

We considered patient-assessed outcomes using region-specific instruments evaluating instability, function, or disability. More specifically, we abstracted data regarding the following patient-reported outcomes: Foot and Ankle Ability Measure (FAAM) [38], Foot and Ankle Disability Index (FADI) [24], CAIT [29], and the Ankle Joint Functional Assessment Tool (AJFAT) [46]. We also recorded adverse effects. We assessed the included outcomes (1) at the end of the rehabilitation period and (2) between 1 and 6 months after treatment.

We proceeded with meta-analysis only for the trials in which the authors performed a baseline measurement using at least one discriminative and/or evaluative questionnaire. The aim of the development of this eligibility criterion was to confirm the diagnosis of ankle instability and/or quantify the functional status of the included participants, thus increasing homogeneity among the included trials in the quantitative synthesis.

Synthesis of the Results

We considered the following arms in the analyses or a combination of them: balance training; strengthening exercises; balance training with sensory-targeted ankle rehabilitation strategies (STARS); destabilization devices; trigger-point dry needling (TrPDN); stretching; manual therapy (ie, manipulation and/or mobilization and/or strain-counterstrain and/or plantar massage); vestibular–ocular reflex (VOR) enhanced balance training; control (sham treatment and/or a wait-and-see policy); and multimodal intervention. As a minimum, multimodal treatment was targeted at three of the following areas of rehabilitation: functional activity, ROM, strength, and balance. We developed this impairment-based group according to the proposed treatment paradigm presented in Donovan and Hertel [18]. To increase the validity of our results, we considered only questionnaires with high responsiveness and test-retest reliability in the analyses [19, 24, 29, 38, 46].

Statistical Analysis

In this systematic review, we performed random-effects pairwise and NMAs using standardized mean differences (SMDs). For pairwise comparisons, we conducted a meta-analysis of change-from-baseline scores using Review Manager (RevMan) software (Version 5.3; Copenhagen, Denmark: The Nordic Cochrane Centre, The Cochrane Collaboration; 2014) [55]. We used STATA software (Release 13; StataCorp LP, College Station, TX, USA) to perform NMA of change scores [47, 57]. We used self-programmed routines and the network command to perform an NMA [7].

We considered all instruments that assessed instability, function, or disability in an enrolled trial (see Table, Supplemental Digital Content 2). For these instruments, we accounted for the direction and maximum possible values of the available scales before calculating the mean outcome values with their SDs.

We visually presented the network of interventions with a modified NMA plot (Fig. 2) [6]. In this plot, the thickness of the edges is proportional to the number of studies for each comparison, and the size of nodes is proportional to the number of participants randomly assigned to each intervention. Dark, light, and dashed gray-colored edges indicate comparisons at low, unclear, and high risk of inadequate allocation concealment, respectively. We also explored the major assumptions in NMA (ie, transitivity and consistency) and used the surface under the cumulative ranking probabilities to rank the efficacy of the included treatments [6, 47, 48]. To explore inconsistency, we used statistical models including but not limited to a node-splitting approach [15, 58]. For this exploration, the results were not statistically significant (p > 0.05), and thus the validity of our results was not compromised.

Fig. 2.

The NMA plot for the assessment of perceived stability at the end of the rehabilitation protocols is depicted. Nodes are weighted according to the number of trials including the respective treatments. Dark, light, and dashed gray edges indicate pairwise comparisons at a low, unclear, and high risk of inadequate allocation concealment, respectively.

We also considered data from controlled trials to assess the presence of small study effects using the Egger’s test and through visual inspection of the comparison-adjusted funnel plot [6, 20]. By small study effects we refer to the occasion in which smaller studies show significantly greater effects compared with larger studies and not the “sparse-data bias,” which refers to a small number of events in some trials [39]. Although the results of Egger’s test indicated that there was evidence of small-study effects (p = 0.009), we were not allowed to draw safe conclusions on the assessment of publication bias in this NMA. This was because the included controlled trials contributed < 50% to the network estimates.

Sensitivity Analyses

We conducted a preplanned sensitivity analysis keeping only trials at a low risk of bias [26]. Only if adequate randomization and allocation concealment were achieved did we judge a trial to be of high quality. After executing the aforementioned prespecified subanalysis on the risk of bias, we noticed that the main findings of this NMA remained robust.

Clinical Interpretation of the Results

We used Cohen’s rule of thumb to classify the effect sizes [9]. Accordingly, an SMD value of 0.2 denoted a small effect, a value of 0.5 indicated a moderate effect, and a value of 0.8 demonstrated a large effect.

We back-transformed SMDs to mean differences with CAIT [29, 52]. Cumberland ankle instability scores vary between 0 and 30, and higher scores represent better ankle stability. For this instrument, we interpreted the results using the minimum clinically important difference (MCID) and minimal detectable change (MDC) of 3 points [52, 59, 62].

Risk of Bias Assessment and Evaluation of the Quality of Evidence

Two reviewers (KT, HSV) independently assessed the risk of bias within and across randomized trials using the Cochrane risk of bias tool [26]. To evaluate quality within trials, we considered the following elements: sequence generation; allocation concealment; blinding of participants, personnel, and outcome assessors; incomplete outcome data; selective reporting; and “other bias.” We judged each entry to be at an unclear, low, or high risk of bias. In this systematic review, 12, eight, and one trials were considered to be at a low, unclear, and high risk of bias, respectively (see Table, Supplemental Digital Content 3). For the quality assessment across trials, we considered the domains of randomization and allocation concealment to be at a low risk of bias. This was because, for the previously mentioned domains, more than half of the information stemmed from randomized controlled trials at low risk of bias.

We used the Grading of Recommendations, Assessment, Development and Evaluations

(GRADE) approach to evaluate the quality of evidence from the networks of interventions [49]. Accordingly, we estimated treatment effects using four levels of confidence (ie, high, moderate, low, and very low levels). We a priori classified the body of evidence to a high-quality rating because only randomized trials were considered. Then, we made judgments about the following elements: inconsistency, study limitations, imprecision, indirectness, and publication bias. We downgraded the quality of evidence by one level as a result of study limitations.

Results

Persistent Instability as Measured by CAIT Scores at the End of the Rehabilitation Period

We found that a 4-week supervised multimodal program was favored over control at the end of the rehabilitation period (four trials, mean difference in CAIT was -10; 95% confidence interval [CI], -16 to -3; p = 0.001) (Fig. 3; see Table, Supplemental Digital Content 4). This difference indicates both a detectable (MDC) and clinically meaningful improvement (MCID) because the score threshold of 3 points in CAIT was exceeded. We also showed that a combined intervention of balance training and strengthening exercises was superior to control (four trials, mean difference in CAIT was -7 ; 95% CI, -13 to -0.1; p = 0.045) (Fig. 3; Supplemental Digital Content 4). Favorable results were demonstrated when the latter combination was supplemented with TrPDN within the lateral peroneus muscle (one trial, mean difference between balance training, strengthening exercises, TrPDN [BT + SE + TrPDN], and control in CAIT was -11; 95% CI, -23 to -0.03; p = 0.050). However, the latter combination was considered in only one trial [50]. For the standalone interventions, NMA demonstrated that a 4-week balance training program was more effective than control in terms of improving CAIT scores, and the difference was clinically meaningful (seven trials, mean difference in CAIT between balance training and control was -5; 95% CI, -10 to -0.03; p = 0.049). Finally, pairwise quantitative synthesis indicated that balance training was favored over control after treatment according to the results of three trials (mean difference between balance training and control in CAIT was -5; 95% CI, -7 to -5; I² = 22%; p < 0.001). Concerning this comparison, the effect sizes were large (Supplemental Digital Content 4).

Fig. 3.

The ranking probability plot for the assessment of perceived stability at the end of the rehabilitation protocols is shown.

In five trials [5, 12, 41, 44, 60], patients were followed up at 1 month and beyond. We were unable to perform statistical pooling because the included interventions could not be connected through an analyzable network of treatments. Limited data from one trial showed that some of the beneficial effects of balance training shown immediately after treatment were maintained at 6-month followup [60]. In this particular trial, the authors proceeded with longer term outcome measurements in regard to a subset of participants of a previously published study [61] and showed that there were no differences between the postintervention and 6-month endpoints.

Reporting of Adverse Effects

Adverse effects were reported in only eight patients treated with TrPDN and proprioceptive/strengthening exercises [50], which included peroneus muscle soreness that spontaneously resolved within 24 to 36 hours after treatment.

Discussion

Clinicians should bear in mind that a large proportion of patients with lateral ankle sprains ultimately develop ongoing instability. Although nonoperative management is indicated as the first-line treatment of ankle instability, a recent meta-analysis underlined that the best approach for nonsurgical rehabilitation intervention has yet to be defined [35]. In the current systematic review, we used an NMA study design, which allowed us to combine direct and indirect evidence to begin to bridge this gap in knowledge. In particular, we aimed to compare patient-reported function in patients with ankle instability who were treated with different nonsurgical rehabilitation interventions because most of these individuals report impaired perceived stability [28]. To achieve this objective, we pooled data from 19 randomized trials and found evidence of moderate strength supporting the efficacy of a 4-week supervised multimodal rehabilitation program. This result remained robust after controlling for the quality of the included studies.

There were several limitations to our study. First, the source studies we included contained only limited reporting of longer term observations. Only five trials assessed the effects of rehabilitation beyond the first month, and these could not be pooled statistically because many different interventions were considered. More specifically, the number of included treatments was greater than that of the studies exploring the results of rehabilitation modalities at 1 month and beyond. We highlight the importance of long-term followup because this has to be accounted for by clinicians in the decision-making process. Second, from a statistical perspective, the large number of treatments—often with few trials on each treatment—limited the power of our results (that is, there was a flat distribution of ranking probabilities). Accordingly, for the interventions that were considered in only one trial each (ie, balance training plus mobilization, multimodal intervention with destabilization devices, stretching, balance training with STARS, VOR-enhanced balance training, and balance training plus strengthening exercises plus TrPDN), the results should be interpreted with caution. Concerning outcome analysis, we combined the results of three different self-administered questionnaires: the CAIT, the FAAM/FADI, and the AJFAT. This diversity of instruments is considered to be associated with introduction of clinical heterogeneity. To deal with this issue, we standardized the results of the included trials before combining them in the analyses as suggested by the Cochrane Collaboration [13]. Finally, it is important to note that only two studies looked at results of rehabilitation interventions at 6 months after treatment [12, 60]. This paucity of studies on longer term observations can be attributed to the difficulties the authors of studies on nonsurgical management of ankle instability have to deal with. To be more exact, substantial loss to followup and inconsistent responses are usually the case in those studies. We recommend authors specifically seek to assess results of these rehabilitation programs in studies that assess patients at longer term followup.

For primary management of ankle instability, the results of the current NMA support the efficacy of impairment-based multimodal programs. To elaborate further, these programs target the major deficits detected in patients with ankle instability and should include strengthening, balance training, ROM exercises, and functional activities. In general, the duration of multistation rehabilitation varies between 4 and 6 weeks [17, 21, 36]. Although an ideal duration has not been specified, a duration of 4 to 6 weeks has been shown to be long enough for improvement of clinical and self-reported ankle function and has been recommended for clinical use [21]. Concerning the quality assessment in this NMA, we found Grade B evidence indicating that the confidence in estimates of effects was of moderate robustness. We also suggest that adherence/compliance to these demanding protocols is crucial. Notably, a 4-week balance training program was the only standalone intervention that provided superior results over control at the end of the rehabilitation period. In other words, if a busy clinician is choosing one of the standalone treatments, a 4-week balance training program has the highest probability of being among the most effective types of intervention. This finding suggests that balance training should be the cornerstone of multimodal rehabilitation protocols targeting self-reported outcomes of function. Balance training interventions can also be supplemented with strengthening exercises to further improve self-assessed outcomes of function. Interestingly, promising results in favor of TrPDN, used as a complementary treatment, were demonstrated. However, because the latter intervention was considered in only one trial, we recommend that future studies explore the efficacy of TrPDN as an adjuvant therapy in rehabilitation programs. Finally, it should be noted that if nonoperative management fails in treating the negative symptoms associated with ankle instability, clinicians should proceed with surgical intervention.

Complications associated with these nonsurgical rehabilitation protocols in general were transient, mild, and uncommon. More specifically, peroneus muscle soreness was observed in a small number of participants. Ideally, a perfect rehabilitation protocol does not involve any adverse effects. However, with certain treatments, short-term complications can be anticipated. The clinician should be aware of the potential for these unintended events and have a recovery plan in place. Common clinical practice for muscle soreness postexercise typically includes a rest period of approximately 24 to 48 hours or the use various therapeutic modalities such as cryotherapy, compression therapy, and massage [43]. When comparing ankle instability rehabilitation treatments, clinicians and researchers need to take into account not only the availability of the patient for rehabilitation, but also how the potential side effects of that therapy may affect their activities of daily living.

The results of this NMA provide robust support that a 4-week supervised multimodal program improves self-assessed outcomes of function in patients with ankle instability. The effect size of the difference between multimodal treatment and control was large. Future studies should evaluate whether short-term benefits of these interventions are sustained over time. Finally, we propose that the authors of future trials on ankle instability comply with the latest inclusion criteria for controlled research to enhance the validity of research in this field [22]. This refers to a minimum set of criteria relating to ankle injury, disability, and function that recently were developed by the International Ankle Consortium, a major international scientific community of clinicians and researchers.