After 35,000 knee operations over 35 years I can reassure that arthroscopic meniscetomy when PROPERLY indicated & performed does help patients. The authors of this paper are welcome to visit us and we will show them how to do the operation properly, then they will be able to publish better results. What on earth are they trying to say "someone in agony with a torn cartilage - Go away, it will fix itself! In my opinion the paper is rubbish.

Greg Keene ISPORTSMED.SA Adelaide, Australia

I agree with Gregory Keene. Unfortunately, in the abstract the exact indication and MRI findings are not defined. But, as I resume from my work with less than 35.000 but many arthroscopic meniscetomies this procedure is effective. Mostly the releaf of hyperextension pain is immediate (back to sports and work). However symptoms in non treated knees improve by the time.To my opinionthe indication for arhtroscopy is a rapid releaf of pain and malfunction by means of a relatively harmless operation. A Gain of 10-20 weeks of life and functional quality.

With all due respect to the vast experiences of individual surgeons, we do know that even the largest single group cohorts fail to achieve the highest levels of evidence.

What we surely cannot negotiate with is evidence, but we can and should criticize study design. As far as I’m concerned, this is a fine study; a well published randomized trial, which will provide substance for tremendous debate, as it is depreciating the value of a very frequent routine procedure.

“Gut instincts, although often correct, can fool us, and should never be leveled with evidence”.

Dear Philipp,

First of all thank you for bringing up always very interesting topics and discussions. I read with much interest the full text article and I’m not surprised with the reported findings.

I think many aspects need to be considered regarding this topic.

My personal experience is much less than 35000 knee surgeries. I’m one of the 4 surgeons responsible for knee surgery at the Orthopaedic and Trauma Center of Torino (University of Torino, Italy). In my hospital 40-50 knee surgeries are performed every week. My colleagues and I, when defining shared approaches to knee pathologies, agreed with a conservative approach with degenerative meniscal tears, whenever possible. NSAIDs, PT and Intrarticular injections are always suggested before sending a patient to the operating room.

As you know, in Italy, we work with a universal healthcare system. Our hospital is highly quoted for orthopaedic pathologies in an area of over 1,500,000 people. As a result, our waiting list for meniscectomy in degenerative tears is over 8 months. By the time the patients are called to schedule their meniscectomy, most of them are asymptomatic.

Regarding the article, I think it is overall well designed. My concern is regarding the lower limb alignment. The authors do not mention AP, lateral, Ronsenberg, Merchant and long leg Xrays. Overload signs (i.e. tibial plateau oedema) on the MRI are not mentioned as well in inclusion or exclusion criteria. Overload of the medial/lateral compartment due to malalignment is the first aspect to exclude in symptomatic degenerative meniscal tears, in order to have happy patients. Regarding this problem, I want to share with you Prof Puddu’s teaching: Whenever a patient with meniscal symptoms does not remember a specific moment related with the onset of the symptoms, always exclude other problems (overload or chondral lesion in malalignment). In other words, when the onset of symptoms is subtle, progressive and not related to a specific trauma or movement, we need to reconsider our meniscal tear diagnosis.

I think there is still room for partial meniscectomy of degenerative tears in patients that need a prompt resolution of the symptoms and in patients unresponsive to conservative treatment. I strongly believe that the conservative option needs to be clearly mentioned during the informed consent. However, the patient ultimately decides.

It would also be interesting to review again the 2 groups of patients in 15 years, and see if there is any difference in terms of arthritis….

Talk about a meaningless paper........

ultimately, we all die, so there is no point doing anything, because the outcome is the same!!!!

I have had arthroscopic surgery to my knee, for a medial meniscus tear. It addressed the mechanical problems, improved my activity level, and my quality of life. I do not know what the outcome would have been at a year, but I certainly would not have been happy to wait to find out. So, is the ultimate point of this paper that arthroscopic surgery is pointless?

It is import to classify

- the cartilage lesions

- the type of meniscal lesions (stable or instale)

- the subcondral bone

to give a correct indication to partial meniscectomy in a degenerative

meniscal lesion

I think this is a very, very important study. As Davide explained, and as the study showed, not every degenerative meniscal tear has to be adressed by meniscectomy. At least the study showed, that meniscectomy may more effective in the first 6 months after surgery. The meniscectomy group had less pain and a better WOMET score in this period of time.

We know that up to 80% of patients receiving meniscectomy will develop an osteoarthirtis after 20 years (today in times of careful partial meniscectomy it is likely to be less). We do not really know how high the osteoarthritis rate is in conservative cohorts, as these patients are usually not followed up that closely.

The most interesting news of this RCT will be likely published in 1-5 years. To see which of the two groups will perform better in the longer run will be exceptionally interesting.

Overall another interesting paper that raises challenge. Like other early interventions changes in the patient profile may not be too significant at 12 months. It is clear that partial meniscectomy of degenerative tears plasy a role in patient care, however these results raise requirements to better understand and define the patient population and explain the risks and benefits of an operative intervention VS conservative treatment.

Like others will be valuable to follow these patients see if there is any difference in terms of development of arthritis and progression to joint replacement surgery.

With all respect to the authors of the study and the editorial board of the journal, we should be very cautious in interpreting the results of a single study that supports an idea or concept that is in contrast to the current clinical practice and our common sense. As we all know, there is despite the possibility of scientific fraud (which I would exclude in this case, but look at historic fake studies etc.) there is a lot of room for bias and error.

First of all, we do not know, how exactly the partial meniscectomy was performed (radical, little resection…, although described by the authors. Several surgeons might have different understandings of the same terms…). I do share the critizism, that there is no information about mechanical axis, joint load etc., patella problems… which probably would have been asked for in an orthopaedic journal…!

The number of patients n=146 is very small, especially for a multicentre study!

The fact, that the authors cannot find a difference between both groups, does not mean, that there isn’t one (type-II-error). This is a frequent misinterpretation in orthopaedic manuscripts.

I have calculated the power of the study results based on the means, standard deviations and sample size provided in Table 1 of the paper with the following results:

Power one-sided power two-sided

Lysholm knee score 0.05 0.05

WOMET score 0.33 0.28

Score for knee pain after exercise 0.22 0.15

Score for knee pain at rest 0.19 0.12

15D score 0.26 0.17

(G*Power Version 3.0.10, Franz Faul, University of Kiel, Germany. Post-hoc analysis, t-test, n1=n2, α-error 0.05)

A power of 0.8 or more usually is considered to be good. This was intended by the authors in an a priori power-analysis. They fail by far to meet an acceptable power of their analysis.

If I take the results for the scores with their SD and calculate an a priori sample estimation, the necessary number of patients in the study would have been: 336914, 381, 699, 965 and 567 in the same order of scores as above!

It is interesting, that the authors do not comment on this, although apparently statistical consultancy was given.

There are some other issues: the patients in both groups were overweight (mean).

A left meniscus lesion does something harmful with the knee: it causes swelling, does mechanically damage the cartilage, activate inflammation cascades and enzymes and cytokines… This is very difficult to measure and is not considered in the sham group. What do the authors think, happended to the torn meniscus?

After successful partial medial meniscectomy I would expect the patient to be in the following condition after one year postoperative: no limping, full weight-bearing, no giving way or locking, no pain, no swelling, no problem climbing stairs, no problem to perform squatting. This “perfect” patient would result in a Lysholm score of 100 points. Noticable pain during heavy exertion, swelling during heavy exertion, little problems climbing stairs and limitation of sqatting not more than 90 degrees would be a less favourable but acceptable result. This patient would still have a Lysholm score of 84 points! Both groups only reach 60 point one year after surgery / sham surgery! Why is that? There are probably other reasons than only the meniscus…

My greatest concern is the last sentence in the conclusion: “These results argue against the current practice of performing arthroscopic partial meniscectomy in patients with a degenerative meniscal tear.” This is not true. As explained above, the authors did not prove, that there is no difference between both groups but instead they were not able to show a difference. To transfer this into the clinical setting, the sentence would be: “These results argue against an attempt to stop performing arthroscopic partial meniscectomy in patients with a degenerative meniscal tear (because the uselessness or inability to show a benefit after one year of intervention could not be shown).” This is a complete different story.

The Editors of the NEJM should be aware of this. I don’t know, why they did not intervene…

This study probably started with good intentions. Now it probably will cause a lot of discussion and misinterpretation of the findings and in my personal opinion, will not help to treat our patients better.

My personal conclusion: although we all can be wrong with “gut feelings”, you can never beat common sense with statistics.

study is interesting but not as interesting as first appears, here's why in my opinion:

patients are those who many surgeons would not operate on anyway

patients had NO mechanical symptoms, no locking

traumatic patients excluded

no osteoarthritis

all they had was knee pain and a degenerate meniscus

no wonder they didn't improve with surgery

an isolated degenerate meniscus has not been linked to pain by any study that I know of, hence implausible from the off

I completely agree with the comment of Jörn Kircher. Also for me, it is difficult to understand the results of this study. Meniscectomy is one of the most common orthopaedic procedures (700'000 per year in the US). How could the editorial board of NEJM accept such an underpowered study with only 146 patients in 5 years? How was it possible that every of the five involved centres included only a mean of 30 patients (one patient every second month)? Are there only low volume centres involved?

This study presumes that meniscectomy in non-traumatic meniscus tears does not help. It is obvious that this cannot be the entire truth. The crucial question for me is: How can we improve our indications to minimize unsuccessful meniscectomies?

I would refer to my own letter to the editor in the NEJM concerning ACL-reconstruction vs. physical therapy (and surgery if required).

http://www.nejm.org/doi/full/10.1056/NEJMc1010531#SA1?url_ver=Z39.88-2003&rfr_id=ori:rid:crossref.org&rfr_dat=cr_pub%3dpubmed

We pointed out that unless the PRO measures used for comparisons are proven to be able to measure differences in the specific patient group(s), then "no difference" is just a non-result. The Lysholm score is not condition-specific for meniscus lesions and the psychometric measurement properties of the WOMET have not yet been confirmed using item response modelling. The authors might perform a post-hoc Rasch analysis on the data from the study in order to address this.

I agree with Jörn Kircher that the study has some limitation and that these limitations are not sufficiently addressed in the paper, but as a pain researcher my attention was drawn to pain outcomes, which favors of the active treatment (figure 2c). This study is underpowered to detect a statistically significant difference and, even more important, the statistical tests could have been better. In their analyses they used t-tests or Mann Whitney. Post hoc adjustment for multiple comparisons will decrease the power and I believe no statistician will recommend using multiple t-tests anymore. Multi level modelling (as MIXED MODEL in SPSS which was the statistical software in use by the authors) has many advantages, among them an increased ability of analyze multiple tests in repeated measurements without loss of power. In addition this statistical method does not require normally distributed outcome measures. Further on, missing values, which is the rule, not an exception in such trials, is handled correctly. The high degree of correlation within each case will decrease the estimated group size when doing power estimation. The take home message from this boring statistical note is that I am not convinced that the intervention in fact was negative, at least when it comes to pain after exercise. It would be interesting to see a follow up of this RCT with improved power and methodology.

Leiv, the data has to be linear for linear models. The data in these studies is response scores to a set of questions (e.g, Does it hurt, yes/no. Yes=1, No=0). The numeric scores are not physical manifestations, but represent a patients perception. The scores are then summed to a total score, and it is assumed that the resultant numeric value means something. If the response scores do not fit an item response function, then it is not linear, and the general linear models are not appropriate. On the other hand, if the data fit a Rasch model (the most stringent IRT model), then the summary scores can be used in any linear model. In other words, if it isnt even measurable, then how can it be measured?

Thank you, Jonathan for an interesting comment. First, pain intensity is not YES or NO but usually 0-100 mm VAS or, as in the paper we discuss, 0-10 numeric rating scale. I agree that it seems rational to use Rasch model when your data is categorical, but repeated measurements of pain intensity can be analysed with mixed model. If your data is non-linear you will have to modify the model but you can still use mixed model,

Leiv, 0-100 mm implies interval data with the unit of pain being millimeters. How can we assume that pain would be a linear entity with categorical data? This is why a probabilistic simple logistic model such as Rasch makes sense, because it models the latent trait as a function of the difference between the level of a persons ability and the difficulty of the item or task to be performed. The model projects a probabilistic Gutman structure, and if the the observed responses to the items converge, then the item set possesses the properties of fundamental measurement. Once this has been confirmed, then it can be argued that future data procured in the same frame of reference (same items and patient type) can readily be analyzed within linear models.

I seem to have gotten off track here and I apologize. My point is that the comparison of groups using latent variable data must be treated with extreme scrutiny, much more than physically manifested data output.

Dear Jonathan. I think we can agree that when testing data sets with repeated measurements for statistically significant group differences, it is best to avoid t-tests, ANOVA and similar methods. I do still advocate using linear mixed model, even for pain intensity since the variable don’t need to be N-distributed and if the development over time is not linear this can be solved by making new time variables. We have modified the model by making time*time and even time*time*time. Testing of interaction between intervention and time (or time*time or time*time*time) is the most important part in the estimates. We have seen that this increase the fit of the model and the estmates fit very well with our observations.

But, the important issue to discuss here is how we should analyze repeated measurements in medicine. Not with t-tests/ANOVA, at least. (I have for some years now wanted to learn more about Rasch analysis. May be you could recommend a course?)

Back awhile MANOVA was often used for repeated measures. I used it in my master project at Lund on knee joint proprioception, but interpreting the results was a beast. It seems Generalized Estimating Equations in SAS is popular these days.

There are two good courses in Rasch here in Europe. Allan Tennant runs a nice course at Leeds which uses the program Rasch Unidimensional Measurement Model (RUMM) to go through the principles and the techniques of Rasch IRT. The course has intro, intermediate, and advanced levels. David Andrich at the University of Western Australia (UWA) offers a great on-line course at intermediate and advanced levels. Here in Copenhagen, there is a course on scale validation that uses Rasch and other IRT models as a base. Svend Kreiner has taught the course, but he has just retired. He probably will keep teaching it together with Karl Bang Christensen and Jacob Bue Bjørner (of SF-36 fame). Leeds is probably the best bet. Let me know if you need more info.

It took 5 centers 6 years to find 205 eligable patients with medial knee pain...

The merit of this study is to confirm that a knee MRI is of questionable value as a selection criterion for meniscal surgery.

Good points Johan. Again, I emphasize that the outcome measures are invalid for the specific condition (chronic degenerative meniscus lesions as confirmed by MRI). Both groups improved, which implies that either sham or active treatment is indicated. This means that sham should be preferred because it is cheaper? The study clearly indicates the use of irrelevant outcome measurement. In Denmark, the head of the Danish Orthopedic Society, a professor of hip surgery, has suggested publicly to drop meniscus operations all together (i.e., traumatic lesions including locked knees etc.). What he doesn't realize is that I can easily put together a study that wont find his surgery more effective than placebo or sham. All I have to do is under-power the study and use an under-validated outcome measure. I was trained as a physical therapist, so I have no interest in showing the virtues of surgery over PT. I do however have a vested interest in performing and pointing out scientifically sound clinical research! The fact that these studies keep getting published in high-impact journals such as NEJM and BMJ is tragic, because it gets picked up by the popular press. From there, it becomes a tool for political manipulation.

With respect to the authors, this is only short-term vision!

Leaving a torn meniscus in place, will destroy the cartilage over time. Unfortunately this will take longer than the follow-up period, mentioned in the paper, but it will and then the knee is lost!

Carefully resecting the torn parts of the meniscus and leaving as much as possible in place will save the cartilage over time.

Hi All,

I perfectly agree with the general idea that the study is well done from a methodological point of view (and therefore rewarded with a publication on the prestigious NEJM), but it's basically inconclusive. As Jörn Kircher said, the study is under-powered and the best conclusion of this research is that the knee arthroscopy (with or without meniscectomy) provide some relief of symptoms at 1 year of follow-up, in people with atraumatic meniscal tears. The study lacks the definition of the location, pattern and size of the meniscal tears, as well as the description of the conditions of cartilage and synovium at the time of surgery. Moreover the study from Katz et al ( cited in this article: ref. n. 8) doesn't conclude only that surgery and physical therapy can provide the same outcomes in symptomatic meniscal tears, but also that there was a 30% of cross-over versus surgery at 6 months of follow-up (35% at 1 year); conversely only 6% of patients assigned to the surgery arm decided to shunt versus physical therapy. In other words, about 1/3 of the patients assigned to the PT arm didn't solve completely their problems only with excercises and preferred surgery.

I'm certain that the knee surgery should be considered with caution in chronic degenerative conditions, without acute trauma, or clear mechanical symptoms (catching or locking), while these studies seem to endorse an opposite idea.

So far, we have to keep in mind that the evidence from the results doesn't fit necessarily with the correctness of the conclusions.

Giovanni, I cannot understand why your points aren't crystal clear to anyone with any research experience. The paper is beautifully written, as was Frobell et. al. comparing ACL reconstruction with PT (and delayed reconstruction), and the study designs are fine. But they are both "non-results". When there is no statistical difference, the best you can say is that you can't say anything. My argument is that if the outcome measures had been confirmed to be valid for the patient group, then you could conduct power analyses that would hold up in a court of research (not law). Sihvonen et al. did conduct a "validity" study of the WOMET for the patients - but they used inadequate methods. We have to remember that with questionnaires, we are trying to measure the patients perception of his/her status - and we know how much perceptions can be influenced by all kinds of exogenous factors. So if the outcome measure is "too fuzzy" to detect differences in the specific population, then this can just as easily be the cause of the result. Stringent validation of patient-reported outcomes was the topic of my doctoral thesis - so I'll admit that I have a certain passion for this subject.

I think the results of the study have given strong statements. Apart from the potential of being symptomatic, a partially torn meniscus is a trigger point for the extension of the tear and ultimately starting to give symptoms. But when will it happen- no body knows because it depends upon the dimensions and the geometry of original partial tear and the level of activity of the patient. So whenever we find a partial meniscal tear in patients who undergo arthroscopy for other knee pathologies, I personally feel that the partial meniscal tear must be balanced and leaving it as such is an actual act of omission which can be a source of future trouble to the patient. The improvement in symptoms in these patients may be due to the irrigation of the inflammatory markers from the joint during the arthroscopy.

Performing a sham surgery is just like removing stones from a diseased gall bladder without doing cholecystectomy with a potential source of the disease still inside.

Please people, this study is not underpowered. It was adequately powered to find the difference they were looking for: that is, a clinically important (around 10 point out of 100) difference. Hence the sample size calculation.

The study IS underpowered to find the ONE point difference that they found, but that is not what they were looking for. They would have been crazy to look for such a meaningless difference.

Simply, there was no meaningful difference between the two groups. This is not "underpowered" and it is not "inconclusive" or a "non-result", it is merely a failure to reject the null hypothesis that there is no (meaningful, clinically important) difference between the two treatments.

And the fact that both groups got better does NOT mean that both treatments were effective. There are many reasons that people get better and having 2/3 or so of your patients improve in ANY study that included a placebo is common (think: natural history, regression to the mean, concomitant treatment, patient expectation ...)

If the patient-reported outcome measure (PROM) doesn't accurately measure what it is supposed to measure, then the power analysis is arbitrary. When we use PROMs to compare treatment effects, we are dealing with psychometrics, where the endpoints are questions about how people perceive their pain, quality-of-life, and function (WOMET). For those questions (or the numeric responses to them) to be useful in a summary score (or total score), which is supposed reflect the "amount" of how much one person or group of persons is affected relative to another, then the validation needs to be non-arbitrary. For example, if I ask these patients to respond to a group of items about how they felt about going through high school, I wouldn't expect to see a between-group difference either. The problem with using correlation, Chronbachs alpha, and exploratory factor analysis to establish validity, is there is no statistical test of confirmation that the items even exhibit valid measurement properties. Valid measurement in terms of PROMs is nicely summarized in a rejoinder written by my statistics professor Svend Kreiner in BMJ rapid response:

http://www.bmj.com/content/346/bmj.f232/rr/637148