Osteoarthritis Prevalence Following Anterior Cruciate Ligament Reconstruction: A Systematic Review and Numbers-Needed-to-Treat Analysis

We performed an exhaustive search of all databases associated with the Web of Science (BIOSIS Citation Index, BIOSIS Previews, Derwent Innovations Index, MEDLINE, and Journal Citations Report) and PubMed databases from 1960 through June 6, 2012. Search terms were osteoarthritis and meniscectomy or anterior cruciate ligament or anterior cruciate ligament reconstruction or anterior cruciate ligament deficient. We included only studies published in English that evaluated the presence of tibiofemoral or patellofemoral OA in patients who sustained an ACL rupture and underwent an ACL-R or had ACL-D. Investigations were included if the presence of OA was assessed radiographically using a standardized assessment tool in either the medial or lateral tibiofemoral or the patellofemoral compartment. A study was excluded if the grade of OA was based upon the contralateral knee or if it was documented that participants had OA at the time of initial ACL rupture. Additionally, research that assessed only the development of OA in participants with rerupture of the ACL graft was excluded. We cross-referenced bibliographies from all relevant studies to identify articles that had not been located during the initial Web of Science search (Figure 1). If the presence of OA could not be determined directly from the study,¹⁹  we e-mailed the authors in an attempt to acquire this information.

Publication bias was assessed for both ACL-R and ACL-D through the creation of 2 funnel plots. The percentage of OA developed in the sample was plotted against the sample size in each study. Funnel plots were then assessed for symmetry, with greater skewness away from the funnel shape indicating a greater degree of publication bias.²⁰ 

From each included study, we extracted the number of patients diagnosed radiographically with tibiofemoral OA and the total number with ACL-R or ACL-D. Similarly, from articles reporting concomitant meniscectomy procedures, we extracted the number of ACL-R and ACL-D patients with OA who underwent meniscectomy and the total number of patients in the ACL-R and ACL-D groups. To determine the progression of OA over time, the data were separated into subgroups based on the decade post–ACL injury in which the radiographs were taken. Studies were partitioned based upon the mean or median time point at which outcomes were assessed; subgroups consisted of the first decade (0–10 years), second decade (11–18 years), and third decade (19–35 years) postinjury. Although Ventura et al²¹  reported mean follow-up times of 19 years post–ACL-R, the range of follow-up times extended into the third decade, so we included their results in the third-decade group to allow ACL-R comparison with the ACL-D study's²²  third-decade OA prevalence data. Results were also separated by study design (retrospective versus prospective), publication year, and graft type chosen for reconstruction to evaluate their effect on the prevalence of knee OA after ACL injury.

The number of patients with knee OA in each group was divided by the total number of participants in each group and then multiplied by 100. Rates from the ACL-R and ACL-D groups were used to calculate absolute risk reduction (ARR; Equation 1) and relative risk reduction (RRR; Equation 2) with associated 95% confidence intervals (CIs).²³  When ACL-R increased the risk of tibiofemoral OA, we reported absolute risk increase (ARI) and relative risk increase (RRI). Absolute risk reduction and ARI were used to calculate the numbers needed to treat (NNT; Equation 3) with 95% CIs (Equation 4).²⁴  Comparisons which indicated a beneficial effect for ACL-R in preventing tibiofemoral OA were termed number needed to treat to benefit (NNT), whereas comparisons depicting an increased rate of OA after ACL-R were reported as number needed to treat to harm (NNH). We calculated odds ratios (Equation 5)²³  and their 95% CIs (Equation 6) to identify the magnitude of risk of incurring tibiofemoral OA depending on the inclusion of patients in the ACL-R and ACL-D groups.

where

ACLR_% = percentage of ACL-R patients with OA;

ACLD_% = percentage of ACL-D patients with OA;

ACLR_#OA = number of ACL-R patients with OA;

ACLR_#NOA = number of ACL-R patients without OA;

ACLD_#OA = number of ACL-D patients with OA; and

ACLD_#NOA = number of ACL-D patients without OA.

Because we were unable to characterize all confounding factors that might have led to the development of OA after ACL injury, we sought to determine if activity levels differed between the ACL-R and ACL-D patients in an attempt to identify possible group differences. The Tegner scale was used to determine the magnitude of change in physical activity level from preinjury baseline levels determined at the time of injury compared with postinjury levels at follow-up. Means and standard deviations for Tegner activity level scores were collected at baseline and follow-up time points. For studies reporting medians and ranges for Tegner scores, we used Equations 7 and 8 for conversion into means and standard deviations, respectively.²⁵  After standard deviations were converted to variance, means and variances for each study were weighted based upon sample sizes from each study. Weighted grand means and variances from baseline and follow-up time points were used to calculate standardized effect sizes (Cohen d) to determine the deleterious effects of ACL-R and ACL-D on the Tegner score (Equations 9 and 10). Preinjury and postinjury weighted means and standard deviations were then used to calculate effect sizes and 95% confidence intervals (CIs) for each group (Equations 8 and 10).

where

m = median;

a = range minimum; and

b = range maximum.

where d = effect size.

Thirty-eight studies^{6,11–17,21,22,26–53}  met the specified criteria and were independently rated by 2 evaluators using the Critical Appraisal Skills Program⁵⁴  to identify methodologic concerns (Table 1). If the evaluators disagreed on the methodologic quality of a study, they discussed their concerns until they agreed on a consensus score. Twenty-seven^* of the included studies specifically evaluated the presence of tibiofemoral OA in patients treated with ACL-R, whereas 2 studies^13,22  assessed OA in only ACL-D patients who received conservative treatment without reconstruction. Additionally, 9 studies^† evaluated the presence of tibiofemoral OA in both ACL-R and ACL-D patient populations. Of the 38 included studies, 12 studies^‡ also reported the prevalence of meniscectomy and the presence of tibiofemoral OA in ACL-R and ACL-D patients, allowing us to evaluate the effect of meniscectomy in ACL-R and ACL-D patients on the prevalence of tibiofemoral OA. One study¹⁹  was excluded because we were unable to determine the presence of OA without a response from the authors. Twelve studies^§ included preinjury and postinjury Tegner scores that were used to assess the change in activity level.

Point estimates and 95% CIs for all outcome measures are reported in Tables 2 through 4. Confidence intervals crossing zero or infinity were interpreted as inconclusive, whereas CIs not crossing zero or infinity were interpreted as conclusive. The mean Critical Appraisal Skills Program score was 9.23 of 12 possible points (range = 7 to 11, mode = 9, median = 9; Table 2). Using the Centre for Evidence Based Medicine guidelines,⁵⁴  which is a rating system assessing both study design and homogeneity of results, we assigned a level of 2a to the data in this review. Looking at the data overall, we assigned a level of 2a based on the homogeneous findings from the cohort studies included in the overall sample analysis (Table 3). Separating the prevalence of OA by time postinjury resulted in heterogeneous findings and a Centre for Evidence Based Medicine grade of 2a−. Similarly, after we stratified by publication year, study design, and graft type, the heterogeneous results received a Centre for Evidence Based Medicine level of 2a−.

The 38 studies consisted of 2837 patients, including 2500 ACL-R patients and 337 ACL-D patients. Overall, patients with ACL-R demonstrated a slightly greater prevalence of knee OA than ACL-D patients, resulting in a moderate NNH with conclusive CIs (Table 3). Additionally, ARI and RRI were low and had conclusive CIs, indicating a minimal increase in risk from ACL-R (Table 3). Odds ratios demonstrated that ACL-R patients were 1.29 times (95% CI = 1.06, 1.52) more likely to have knee OA compared with the ACL-D cohort. Data pertaining to the first decade after ACL rupture yielded a very small increase in OA development in ACL-R patients over those with ACL-D patients (2%); however, the percentage of OA in ACL-D patients was 34% greater than in ACL-R patients in the third decade (Table 3). Confidence intervals for ARR and RRR in the third decade were wide but did not cross zero (Table 3).

Fifty-two percent of patients who underwent ACL-R and a meniscectomy demonstrated knee OA. Of patients who underwent meniscectomy and remained ACL-D, 59% developed knee OA (Table 4). The NNT for ACL-R with meniscectomy was moderate, with inconclusive CIs (Table 4). Additionally, ARR and RRR were low, and the CI for ARR was inconclusive (Table 4). Odds ratios demonstrated that ACL-R patients who underwent a concomitant meniscectomy were less likely to have knee OA than patients who underwent a meniscectomy but had ACL-D (OR = 0.75; 95% CI = 0.31, 1.21). Patients who had ACL-R along with meniscectomy continued to show a decrease in OA in the third decade compared with ACL-D patients who had a meniscectomy, resulting in a small NNT and moderate ARR and RRR (Table 4).

Overall, the prevalence of OA was higher in patients with isolated ACL-R compared with isolated ACL-D patients. For isolated ACL-R, the NNH was small and the CIs were conclusive (Table 5). Moderate ARI and RRI with conclusive CIs were noted for patients with isolated ACL-R (Table 5). Those who underwent an isolated ACL-R were 1.73 (95% CI = 1.44, 2.02) times more likely to have OA than patients who had ACL-D. Small to moderate NNH, ARI, and RRI with inconclusive CIs for ARI and RRI were seen 1 and 3 decades after ACL injury (Table 5).

The numbers of ACL-R and ACL-D patients were similar between retrospective and prospective designs, but ACL-D populations were significantly smaller than ACL-R populations (retrospective ACL-R n = 1455, ACL-D n = 213; prospective ACL-R n = 1099, ACL-D n = 80). Between prospective and retrospective studies, results conflicted. Prospective studies demonstrated that ACL-R patients had an increased rate of OA compared with those who remained ACL-D; NNH was small and CIs were conclusive (Table 6). Conversely, retrospective studies demonstrated that ACL-R patients had a decreased rate of OA compared with those who had ACL-D; NNT was moderate and CIs were inconclusive (Table 6).

We divided the study results into 3 subgroups based upon publication year (1990s, 2000s, 2010s). Only 1 study was published in the 1990s, and it assessed a small number of ACL-R patients (n = 114). For studies published between 2000 and 2009, there were significantly more ACL-R patients (n = 1105) than ACL-D patients (n = 199). A 1% ARR in OA presence was seen in the ACL-R patients compared with ACL-D patients; NNT was large and the CI crossed zero (Table 7). For studies published between 2010 and 2012, there was an increase in the presence of OA in patients who underwent ACL-R, yielding a moderate NNH with conclusive CIs (Table 7).

A total of 1285 patients underwent patellar tendon autograft reconstruction, 476 underwent hamstrings tendon autograft reconstruction, and 125 underwent an open patellar tendon reconstruction. Open reconstruction using a patellar tendon graft resulted in the greatest risk of OA, and the NNH was small (7). Similarly, the arthroscopic patellar tendon autograft had a high OA prevalence, with a small NNH (10), and a conclusive CI (Table 8). Reconstruction using a hamstrings autograft demonstrated a decrease in the presence of OA when compared with the ACL-D group; the NNT was small and the CI was conclusive. However, ARI and RRI were moderate with conclusive CIs (Table 8).

Both ACL-R and ACL-D had a strong negative effect on the level of physical activity from the preinjury time points to follow-up, reflecting decreases in activity level (ACL-R d = −0.90, 95% CI = −0.77, −1.133; ACL-D d = −1.13, 95% CI = −0.96, −1.29).

It is estimated that 12% of the United States population has diagnosed knee OA.⁵⁶  For patients with ACL injury, however, the OA percentage is much greater than that for the overall population and can be as high as 60% to 90%.¹⁶  Our results provide further evidence of a similarly high prevalence of OA after ACL injury. In the total sample, the percentage of patients with knee OA was slightly increased in patients who underwent ACL-R (44%) compared with those who had ACL-D (37%), demonstrating an RRI of 7%. Specifically, those patients with patellar tendon reconstruction technique demonstrated the highest prevalence of OA, with minimal difference in OA between open (49%) and arthroscopic (47%) procedures. However, patients who had ACL-R with a concomitant meniscectomy displayed a definitive reduction in OA prevalence compared with those who remained ACL-D and underwent a meniscectomy.

In the first decade after injury, the ACL-R group continued to show a greater prevalence of OA (36%) than the ACL-D group (34%), with this trend continuing into the second decade after ACL injury. In the third decade after injury, ACL-D patients had a significantly greater rate of OA compared with ACL-R patients, but only a small cohort of ACL-D patients was assessed during this time. These results seem to suggest that ACL-R may decrease the likelihood of maintaining long-term joint health compared with ACL-D. Additionally, it should be noted that decrements in physical activity were similar between ACL-R and ACL-D cohorts. Although we cannot identify other confounding differences between cohorts that might have led to OA, the reductions in physical activity were likely not a factor.

Similarly, the subset of patients with isolated ACL-R and no concomitant meniscectomy were 1.73 times more likely to develop OA than ACL-D patients with an isolated ACL injury. Our results demonstrate that ACL-R was not an adequate prophylactic intervention for decreasing knee OA development after an isolated ACL rupture. However, it is possible that these patients sustained meniscal damage that did not require a meniscectomy and was not reported by the authors. Anterior cruciate ligament reconstruction is performed at a rate of approximately 250 000 cases per year.²  The cost of ACL-Rs, including diagnosis, surgery and rehabilitation, is around $17 000, with a total annual cost of approximately $3 billion in the United States.⁴  These data demonstrate the low potential that ACL-R offers prophylactic benefit in maintaining long-term joint health. The entire data set demonstrated an NNH of 16, suggesting that for every 16 patients that undergo ACL-R, 1 more patient will develop OA compared with the ACL-D group. This relatively low NNH indicates the increased risk for developing OA after reconstruction compared with remaining ACL-D.

When an ACL injury occurs in combination with meniscal damage warranting a meniscectomy, performing an ACL-R along with meniscectomy may be beneficial for reducing the risk of OA. In meniscectomy patients, ACL-R and meniscectomy decreased the rate of knee OA by 7% compared with ACL-D patients who received a meniscectomy; however, the CI for ARR crossed zero. This trend continued in the third decade postinjury, with ACL-R and meniscectomy patients exhibiting 61% less OA than patients with ACL-D and meniscectomy. However, it should be noted that results in the third decade postsurgery were extrapolated from only 1 study²²  that had a small sample size of ACL-D patients at the farthest time from injury (35 years). This may be further evidence that the meniscus is a vital component of maintaining joint health and that removing the meniscus can decrease force transmission and increase degeneration of bony tissue,¹⁸  which may be hastened by an unreconstructed ACL rupture. Isolated meniscectomy has previously been shown to increase the risk of developing knee OA.⁵  It has been suggested that removing a portion of the meniscus may decrease the ability to attenuate energy within the knee joint.¹⁸  Additionally, partial meniscectomy has been associated with decreased quadriceps strength, which may alter lower extremity biomechanics and complicate the already decreased ability to attenuate energy at the knee.⁵⁷  We found that in patients with ACL rupture and meniscal damage, ACL-R and meniscectomy demonstrated greater prophylactic benefits in decreasing knee OA than in those with ACL-D who underwent meniscectomy.

Isolated ACL-R increased the risk of knee OA compared with ACL-D patients within the entire population as well as in each decade after injury. This increased risk of OA may be evidence that although ACL-R is effective in restoring joint stability after injury, it may not be the most efficient treatment in decreasing the development of posttraumatic OA when the ACL is the only involved structure. An NNH of 9 suggests that for every 9 patients undergoing ACL-R, 1 additional patient will develop knee OA compared with the ACL-D group. Although ACL-R is common for much of the population incurring ACL injury in the United States, it is not the standard of care in countries other than the United States.⁵⁸  Early reconstruction may not be more advantageous in improving patient-reported outcomes compared with conservative treatment with delayed reconstruction.⁵⁹  Therefore, conservative treatment for patients with isolated ACL ruptures may be prudent before deciding whether to reconstruct the ligament. More patients with isolated ACL ruptures may be able to cope without an intact ligament. Many factors enter into the decision to undergo ACL-R or conservative treatment, and the risk of OA development should be a major concern. If isolated ACL-D patients are able to achieve their desired physical activity levels and accomplish activities of daily living, they may benefit from conservative treatment rather than ACL-R. Copers are patients who can remain ACL-D and return to premorbid levels of physical activities without reconstruction.³  Noncopers are not able to return to premorbid levels of activity and are likely to benefit from ACL-R to regain normal arthrokinematics before returning to full activity.³  It may be more cost-effective to screen and treat noncopers with ACL-R as a matter of ACL injury management. Recent evidence⁶⁰  has suggested that the functional status of the knee after ACL rupture may indicate if a patient is a better candidate for repair than conservative treatment. However, most young athletes forego an initial bout of conservative treatment in order to immediately regain ligamentous stability of the knee and return to activity as soon as possible. Determining if a patient can cope with ACL-D may save health care costs, reduce time spent recovering from surgery, and prolong long-term joint health after injury.

Although ACL-D patients demonstrated a greater decrease in physical activity than ACL-R patients after injury, the 2 were not significantly different from each other because their CIs overlapped. Because this decrease in physical activity was statistically similar among ACL-R and ACL-D patients, altered levels of physical activity may not be the most critical factor to explain the development of OA after ACL injury with similar risks in reconstructed and deficient patients.

Most studies had a wide range of patient populations, ages, activity levels, and follow-up times, but not all authors reported all demographics,^{11,16,17,26,29}  which made comparisons among studies difficult. Because of the differences in study design and outcomes assessed in individual studies, we were unable to determine if all potential OA risk factors were similar between ACL-R and ACL-D patients. Although not knowing the equivalency of risk between groups remains a limitation to the current study, we found that each group displayed a similar magnitude of decreased activity. The number of ACL-R patients was significantly greater (n = 2500) than the ACL-D patients (n = 337), which may reflect the bias in standard-of-care practices and might have affected the variability in our calculations. In addition, a very small number of patients were assessed during the third decade after ACL injury. This could indicate a lack of information about patient outcomes beyond the first 2 decades postinjury or possible publication bias against the studies with a follow-up greater than 20 years. Different ACL-R procedures could lead to differences in OA development, as it has been reported⁶¹  that graft types affect patient outcomes. Anterior cruciate ligament reconstruction with a hamstrings tendon graft has demonstrated better self-reported function and decreased anterior knee pain compared with reconstruction with a patellar tendon graft.⁶²  In a recent study by Frobell et al,⁶³  patients reconstructed with a patellar tendon graft demonstrated a significant increase in patellofemoral OA but not tibiofemoral OA compared with patients who received hamstrings tendon grafts, suggesting that graft choice and the potential effects on long-term joint health should be considered when patients elect to undergo ACL-R. With regard to the relationship between OA development and the type of graft used for reconstruction, hamstrings autografts seem to be associated with lower OA prevalence and lower NNT than other graft types, which may indicate the importance of graft selection for patients undergoing reconstruction.

Also, recent advancements in the ACL-R procedure may lead to different outcomes than the older procedures used in the reviewed studies. Publication dates ranged from 1991 through 2011; therefore, a large range of techniques was likely used in these studies. Articles published after 2010 demonstrated more OA in patients undergoing ACL-R than in ACL-D patients compared with articles published between 2000 and 2009. However, the studies published between 2010 and 2012 had small cohorts of ACL-D patients (n = 58) and also tended to have longer follow-up times (mean = 13.6 years) after injury, during which the rate of OA would be expected to increase. Radiographic OA grading scales were also not consistent across all studies (Table 1); however, we were not concerned about the grade of OA in this study but rather the presence of the disease. Authors of all included studies diagnosed OA through radiologic examination. Although OA was graded using different scales, the criteria for determining OA were consistent across all scales.

Both ACL-R (Figure 2) and ACL-D (Figure 3) studies had the potential for publication bias, as shown by the skewness of the funnel plots. Skewness of the plots may be due to small sample sizes or participant-recruitment processes.

The demonstrated publication bias suggests that investigations with results yielding greater increases in OA development after ACL-R may be less likely to be published. Specifically, the ACL-R funnel plot demonstrates that the studies with the greatest sample sizes were to the right of the mean. This may indicate that even though these studies had the greatest sample sizes, they did not represent the entire ACL-R population. Furthermore, authors of 2 of the studies^26,30  with the highest numbers of participants did not report patient demographics, which could have led to selection bias if only the patients with the best outcomes were included. The ACL-D funnel plot identified 1 potential outlier²² ; according to that report, 98% of patients developed knee OA 35 years postinjury. This study had a small sample size (n = 19) and evaluated outcomes at the longest time postinjury, thus potentially reflecting publication bias in indicating worse OA outcomes after ACL-D. More than half of the included articles were retrospective in nature, which may have led to greater variability in the measures of OA development after ACL injury. Retrospective designs showed a decreased risk of OA development in patients undergoing ACL-R compared with the ACL-D group. Conversely, prospective designs showed more OA in patients undergoing ACL-R than in the ACL-D cohort, with an NNH of 4.

Furthermore, we were unable to identify the individual factors that resulted in a patient's choice to reconstruct or not. Another factor that may influence outcomes is the effectiveness of nonsurgical rehabilitation after injury. It is difficult to determine if OA development is mainly explained by factors related to the surgical intervention or if the nonsurgical rehabilitation performed after ACL-R and ACL-D is critical in obtaining optimal therapeutic outcomes.

In conclusion, the current literature does not provide substantial evidence to suggest that ACL-R is an adequate intervention to prevent knee OA. Overall, the risk of knee OA was slightly higher in patients undergoing ACL-R than in the ACL-D group, with ACL-R resulting in an NNH of 16. Isolated ACL-R increased the risk of OA, yielding an NNH of 8. With regard to OA prevalence, the only patients benefiting from ACL-R were those undergoing concomitant meniscectomy with reconstruction, compared with ACL-D patients undergoing meniscectomy, with an NNT of 15; however, the CIs were inconclusive. Although both populations demonstrated decreased activity levels, they did not differ from each other, showing that decreased activity level may not be a confounding factor for OA development when comparing ACL treatment options. These analyses may provide evidence for clinicians working with young athletes that the long-term goal of prolonging joint health after ACL injury is of critical importance.