STABILITY 2: Anterior Cruciate Ligament Reconstruction +/- Lateral Tenodesis With Patellar vs Quad Tendon

Anterior cruciate ligament reconstruction (ACLR) is complicated by high failure rates in young, active individuals, which is associated with worse outcomes and higher rates of osteoarthritis (OA). ACLR failure reduces quality of life (QOL) and has substantial socioeconomic costs. Therefore, strategies to reduce ACLR failure are imperative. Lateral extra-articular tenodesis (LET) may provide greater stability; however, its effect on the rate of graft failure remains unclear, and surgically-induced lateral compartment OA is a concern given the potential for over-constraint of the joint.

Many surgeons believe that autograft choice for ACLR, with or without LET, does not affect graft failure. Specifically, bone patella tendon bone (BPTB) autograft has been perceived to be just as good as a hamstring tendon (HT) graft. However, recent meta-analyses suggest that BPTB grafts provide better stability, albeit with greater donor site morbidity. Increasingly, quadriceps tendon (QT) autograft is being used for ACLR with claims of comparable stability to the BPTB graft without the donor site morbidity. However, the effects of a QT on graft failure are unknown. Despite its importance, there has not been an adequately powered study to evaluate if BPTB or QT is superior to the other in terms of graft failure rates, return to sports, donor site morbidity, lateral compartment OA and healthcare costs.

Objectives:

Determine if graft type (QT, BPTB, HT) with or without a LET affects:

• Rate of ACL clinical failure 2 years after ACLR;
• Patient-reported outcomes, muscle function, performance-based measures of function (hop tests, drop vertical jump) and return to sports;
• Intervention-related donor site morbidity, complications and adverse outcomes;
• Cost-effectiveness of ACLR and LET. Approach: This is a multicenter, international, randomized clinical trial that will randomly assign 1292 ACL deficient patients at high risk of re-injury, to an anatomic anterior cruciate ligament reconstruction (ACLR) using a BPTB or QT autograft with or without a LET in a 1:1:1:1 ratio. Data from this study will be combined with data from a recently completed randomized clinical trial comparing ACLR with a hamstring tendon (HT) graft with or without LET.

Randomization will be stratified by surgeon, sex, and meniscal status (normal/repaired v meniscectomy) in permuted block sizes to ensure that any differences in outcome attributable to these factors are equally dispersed between treatment groups. Each site will either use traditional or expertise-based randomization. All randomization will use the web-based application available through the data management center.

Methods to Reduce Biases:

Selection Bias between STABILITY 2 Intervention Groups: We will partially determine eligibility prior to surgery. Once in surgery, all patients will undergo an examination under anesthesia and diagnostic arthroscopy to confirm final eligibility. The surgeon will document evidence of the participant's ineligibility in the surgical report that is discovered during surgery (e.g. partial ACL rupture where an ACLR is not performed, multiple ligament reconstruction, chondral lesion requiring more than debridement). The operative notes for all participants that were consented will be included in the study database. The study quality control monitors will review the evidence provided by the operating surgeon (arthroscopic pictures/video of ACL integrity and chondral status) and recommend that either the participant remain in the study or be withdrawn since they were never eligible.

At the traditional randomization sites, full randomization occurs during surgery following arthroscopic evaluation of eligibility, which already serves to reduce the risk of selection bias. The action of requiring evidence of ineligibility at time of surgery therefore, reduces the risk of sampling bias (applicability) in traditional randomization sites. At the expertise-based randomization sites, where randomization to graft type occurs prior to surgery, this action will prevent unsubstantiated post-randomization withdrawals prior to randomization to LET or no LET, which reduces sampling bias (applicability) and selection bias by avoiding unequal exclusions between the LET/no LET assignment since randomization to LET/no LET occurs after the arthroscopic examination. In summary, having to provide evidence of eligibility at surgery will serve as a deterrent for surgeons declaring eligible consenting patients ineligible during surgery, which serves to reduce the likelihood of sampling and selection bias.

Selection Bias between STABILITY 1 (NCT02018354) and STABILITY 2 Comparisons: STABILITY 1 followed the exact same protocols as are proposed for STABILITY 2 and the two studies will be performed immediately in series; thus, changes in ancillary care and surgeon expertise are unlikely. Consequently, analyses that combine data from STABILITY 1 and STABILITY 2 are unlikely to suffer significant between-study selection biases that are usually a concern for non-randomized comparisons. Further, to evaluate selection bias between the STABILITY 1 and STABILITY 2 samples, the baseline characteristics of the samples will be evaluated to identify any systematic differences between the samples.

Performance Bias, Fidelity \& Adherence: Surgeons have agreed upon standardization of aspects of the surgical interventions that could potentially influence outcomes. All other aspects of the surgical interventions are meant to be pragmatic and may vary by surgeon. Aspects allowed to vary are not expected to influence outcome. Further, randomization is stratified by surgeon so that nuance differences by surgeon are balanced between groups. In terms of fidelity, all participating surgeons have the necessary expertise to conduct both surgical procedures (BPTB, QT) if they have elected to participate in traditional randomization. Surgeons who have a preference for or greater skill performing one graft type over the other, will participate in expertise-based randomization and have identified another surgeon with similar expertise/preference performing the opposite graft type. In terms of performing a LET, all surgeons who have not completed at least 10 LETs will participate in a cadaver training lab and be required to complete at least 10 LET procedures prior to randomizing their first patient. The investigators have agreed upon a protocol for ACL rehabilitation following ACLR. All patients will receive a copy of the protocol with a standardized referral from their surgeon for their physical therapist. Deviations from the protocol are not expected to be different from usual practice and as such patient adherence with rehabilitation protocols is expected to vary. Given the large sample size, we expect that adherence to rehabilitation will be balanced between groups and we will adjust the analyses for length of time in rehabilitation. This study will track the number of rehabilitation sessions attended, milestones and timing of rehabilitation-specific activities to collect some adherence and fidelity data.

Detection Bias: An independent surgeon, primary care sports medicine physician, physical therapist or athletic trainer who is unaware of group allocation will conduct all assessments of graft stability (primary outcome). Although incisions are unique for each procedure, patients will wear a tubigrip sleeve over both knees to conceal the incisions and reduce bias in assessments that require side-to-side comparisons, including the primary outcome. Data assessors for other outcomes will also be kept unaware of group allocation using this method.

Intention-to-Treat Principle: Patients will be analyzed within the group to which they were randomized regardless of graft type received or adherence to protocols.

Attrition Bias: From STABILITY 1, we have complete data on 95% of the 618 patients who are at least 2 years postoperative demonstrating that we are capable of successful recruitment and retention in a study of this magnitude. We will use the same measures to maximize completeness of follow-up

Statistical Methods:

Sample Size: The absolute risk of clinical failure of the ACLR (as defined by either graft rupture/symptomatic instability requiring revision ACLR surgery or persistent rotational laxity as measured by an asymmetrical positive pivot shift compared to the contralateral side) is estimated to range from 25-35%. We consider a relative reduction in the rate clinical failure of ACLR of at least 40% by 24 months after surgery to merit a change in practice (i.e. of sufficient magnitude to warrant the additional costs of adding a LET). Since, our primary interest is in determining the main effect of graft choice and whether the effect of LET varies by graft choice, the focus will be on the following comparisons: 1) HT+LET versus HT (already shown by STABILITY 1), 2) BPTB+LET versus BPTB, 3) QT+LET versus QT, 4) BPTB versus HT+LET, 5) QT versus HT+LET, and 6) BPTP+LET versus QT+LET.

With 210 patients per group and a type I error rate of 1%, we would have 80% power to detect a hazard ratio of 0.56 (i.e. 44% clinical failure risk reduction when comparing the LET v no LET condition) assuming the clinical failure rate is 33% (the average rate of failure in STABILITY 1). A small type I error rate of 1% was used to reduce the risk of a type I error due to the multiple comparisons based on the Bonferroni method to achieve an overall type I error rate of 5%. Even if there is an intra-cluster/surgeon correlation coefficient (ICC) as large as 0.02, 1) the average number of surgeons per site is 3 given the number of surgeons at each site ranges from 1 to 4, and 2) the average number of patients per surgeon is 22, we will need 281 patients per group to account for the clustering effect. To reduce the risk of losing precision from patients withdrawn and lost to follow-up, an additional 15% of patients will be recruited (attrition was 5% from STABILITY 1), for a total of 323 participants per group or 1292 patients total (or 1910 when STABILITY 1 and STABILITY 2 data are combined).

For the Sex-based Research Question

Preliminary results of STABILITY 1 suggested that HT+LET is superior to HT ACLR alone and is associated with an increased odds of failure compared to HT+LET for both males (odds ratio (OR) = 2.53, 95% confidence interval (CI) = 1.42, 4.51) and females (OR = 1.76, 95% CI = 1.05, 2.96). Given these results, and because females tend to be quadriceps dominant in their landing biomechanics compared to males, and use of a HT graft is currently the most common method of ACLR, we need to understand whether harvesting the HT (which may further contribute to quadriceps dominance) should ever be a first-line option for females. Thus, STABILITY 2 will compare failure between HT+LET and other graft options (BPTB or QT) for males and females separately.

Among the 309 patients per group, we assume half will be female (51.5% of STABILITY 1 participants were female). Thus, we expect to have 159 females in each of the HT+LET, BPTB, and QT groups. Given the failure rate of 29% for females when treated with HT+LET (based on the result from STABILITY 1), the minimum detectable OR will be 2.1 with a power of 80% at the significance level of 0.05. Given the failure rate of 21% for males when treated with HT+LET (based on the result from STABILITY 1), we will be able to detect an OR of 2.4 with a power of 80% at the significance level of 0.05. According to the rule of thumb on magnitudes of effect sizes by Cohen, an OR of 2.1 or 2.4 is considered to be as small (1.5) to medium (3.5) effect size, i.e. with the sample size of 309 (159 females and 150 males) per group, we will have a power of 80% to detect a small to medium treatment effect for males and females separately at the significance level of 0.05.

Statistical Analyses: The data collected through this study will be pooled with the data from STABILITY 1 for analysis (n=1800). To determine whether graft type (QT, BPTB, HT) with or without a LET offers a greater reduction in rate of failure following ACLR (primary research question), we will use a random-effects logistic regression with failure following ACLR at each visit (yes/no) as the outcome where fixed effects include intervention group, meniscal repair status, sex and time (as a categorical variable) and random effects include patient and surgeon. We will conduct a similar analysis for secondary outcomes like return-to-activity and donor site adverse events, as both are binary outcomes. For each continuous secondary outcome including patient-reported outcomes (PRO) scores, measures of impaired range of motion (ROM) and muscle strength, performance-based measures of physical function, and lateral compartment joint space narrowing, we will conduct a linear mixed-effects model where the fixed effects include ACLR group, meniscal repair status, sex and time (as a categorical variable) and random effects including patient and surgeon. For missing data, we will evaluate whether data are missing completely at random by comparing the available data (especially at baseline) for those with and without missing data at follow-up. We will use multiple imputation techniques to handle missing data.

Sex-based analysis: To compare failure between HT+LET and other graft options (BPTB or QT) for males and females separately, we will conduct a random-effects logistic regression with the same fixed and random effects as in the primary analysis.

Health services analyses: We will assign the average procedure cost for an ACLR surgery at each participating institution with the additional cost of the lateral extra-articular tenodesis for those patients randomized to the LET group. Patients who undergo a revision ACLR will complete a healthcare resource diary to capture additional direct and indirect costs. We will conduct a cost-effectiveness analysis from a healthcare payer and societal perspective using quality-adjusted life years (QALY) as our effectiveness outcome at two years postoperative. We will estimate the incremental net benefit (INB) of ACLR + LET using a random effects multilevel model. To characterize the statistical uncertainty around our estimate of INB, we will use an extension of the standard net benefit regression framework using the hierarchical data to generate location-specific net benefit curves, and cost-effectiveness acceptability curves.