Here is a final installment in our coverage of the Middlekoop paper. First up, we had Neil O’Connell talking about elephants and then we had Peter O’Sullivan raising some provocative thoughts on the value of our current direction in trying to evaluate exercise as a treatment for back pain. Now, from that odd group of people called Bio-statisticians, comes an important consideration that may leave us all scratching our heads…
From Dr Anne Smith
The validity of any meta-analysis is questionable here even under a random effects model such as the one used in the Middlekoop paper[1], as heterogeneity between studies is likely to be very high, many of the studies were at high risk of bias (60%), and in all analyses the number of trials included was only 6 or less, and in most cases only 2 or 3. Continuing on with these kind of meta-analyses is futile at the present time. However, it is unlikely that we will ever reach the situation in which we are able to pool results from large, homogenous trials with minimal bias in the area of CLBP. Indeed it is impossible to eliminate some aspects of bias from these sorts of trials. Here are some questions I think are important:
(1) How do we effectively blind patients from their intervention and thus minimise effects of what are often very powerful belief systems about what particular form of exercises will help them?
(2) How do we blind care providers so they do not bring their belief systems into the mix?
(3) How do we absolutely ensure patients are only exclusively receiving a particular exercise intervention and not other co-interventions?
It is only human for care providers to want to assist their patients as much as possible, and so subtle co-intervention bias is likely to always be present. Surely it would be better at this stage to put our time and resources into expanding our understanding of the many pathways to chronic and persisting pain and disability, and then on this basis design targeted interventions.
From Lorimer:
Nice work Anne – don’t hold back will you. I would like to float a couple of answers to those excellent questions. We have tried to address the first one by telling patients in both groups that they are in the active treatment group that is being compared to a control treatment. We have tried to address the second one by telling them the same thing. We also measure as much of these things as we can and control for them in statistics – VAS on expectations of patient and practitioner, beliefs and attitudes of both, enthusiasm and perceived knowledge level of practitioner. Intriguingly, reviewers of these papers often suggest they are meaningless data and should not be included. I agree with Anne that they are not meaningless – quite the contrary – I wonder if they are very VERY meangful.
About Anne
Anne Smith is a really nice, totally understated and very clever person who has advanced training in bio-statistics (a language that few of us understand). She originally trained as a physiotherapist, but because of her frustration with current clinical practice, she embarked on a research career. She is currently a post doctoral fellow at Curtin University, Perth, Australia (in spite of the fact that she could be working in the private sector earning squillions with her advanced stats skills). She has a big knowledge and interest in the complexities of chronic spinal pain disorders, and the pathways that lead to their development. She has published lots of great papers and does really clever things with complex data sets, to answer clinical questions in order to tell a clear story. She actually reads stats modelling books on holiday because they interest her (that is, if she is not surfing, running with her black labrador or hanging out with her family). Clearly this bio was not written by her.
[1] van Middelkoop M, Rubinstein SM, Verhagen AP, Ostelo RW, Koes BW, & van Tulder MW (2010). Exercise therapy for chronic nonspecific low-back pain. Best practice & research. Clinical rheumatology, 24 (2), 193-204 PMID: 20227641
All blog posts should be attributed to their author, not to BodyInMind. That is, BodyInMind wants authors to say what they really think, not what they think BodyInMind thinks they should think. Think about that!
{ 5 comments… read them below or add one }
Hi Ann,
I totally agree that calculating pooled effect sizes from such a range of studies is questionable and the bias problems are very real but even accepting such imprecision exercise doesn’t look great across the trials particularly as the weight of bias tends to favour the “active” arm.
I love the idea of looking closely at the pathway to chronicity but what amazes me across the existing big prospective observational studies in back pain is the lack of decent clues. Again methodological issues get in the way of clarity but for non-specific back pain the predictors of outcome are not great apart from things like initial levels of pain and disability. Psychosocial factors like distress, depression, self efficacy come in and out of focus despite no shortage of data. I wonder if we could ever get this data homogenous and convincing enough to drive better clinical research/therapy provision?
[Reply]
Since Butler, Moseley, and the like have filled my head with all of their strange thoughts, I am less surprised that the spinal pain research has been less than stellar. All of the interventional techniques have focused on changing nociception, but the measurement tools have been instruments to measure pain, function, etc. Why wouldn’t we expect a poor correlation? They are not the same. One would expect better grapes to produce better wine, but that is not true. A lot can go on between the vine and the bottle on my table that can confound that correlation.
[Reply]
Why am I not surprised that one can bring exercise in back pain round to wine? Nice work John.
[Reply]
Just throwing some questions out there… Do we as humans make the issue more difficult than it really is? For example, Lee Goldman worked with some mathematicians interested in creating statistical rules. Goldman was interested in knowing what factors predicted a heart attack, so he dumped boatloads of patient data into a system and came up with an algorithm. Mathematically, he came up with an algorithm in how to treat chest pain. Obviously, chest pain can be much more deadly than back pain (unless of course one drinks wine in moderation which does have a positive effect on the heart… but that’s another topic). Goldman defined combinations of risk factors and created decision trees for treatment options. Goldman’s algorithm guessed right something like 95% of the time. Goldman made the process simple and basically took out a ton of individualistic factors because those factors were just “noise.” Humans thought a lot of factors were relevant and important, but reality they weren’t. Treatment decisions were based on ECG results and 3 other factors – that was it. (Now we won’t have a discussion on how cardiologists felt about using mathematics to make clinical decisions… that’s a whole different topic.)
We have a boatload of back pain virgins (first-timers/first episode of back pain) all around the world. Neil mentioned some factors that need to be collected and dumped into some system… there are other factors… and mix in some examination findings too. I doubt there will be a true homogeneous group, but I’d be willing to bet, that mathematically, certain factors could shake out to help determine effective treatment interventions. In other words, there might be a way to create a decision-tree for low back pain but NOT when the person is in the chronic state… when the patient first has an episode. I am under the assumption if the right intervention is provided at the right time, there is a better chance at reducing the frequency of chronic back pain.
Would it even be possible to create a list of factors mathematically derived to help highlight what might be common for people with chronic low back pain (we have a boatload of that population too)?
When we see what the factors in the chronic back pain situation are and the strength of those factors is there a way to change the course to prevent those factors by implementing them in the decision-making tree during the first episode of back pain?
In other words, I think we need to know if factors in the chronic back pain situation are also potentially present in the acute back pain situation. And then, to really add confusion… if factors in the acute back pain situation are different than in the chronic back pain situation – when and why do factors change?
[Reply]
More from author
Some great points raised in this discussion.
The comment by Neil expressing doubt that we will ever get data from longitudinal trials homogenous and convincing is totally understandable.
And it is also true that we probably consider way too much information and therefore struggle to find the true keys to focus on. Goldmans algorithm uses a technique called ‘recursive partitioning’ which is a really nice way to try and get a handle on all the complex interactions on your data and simplify things down to a few key variables. When we are trying to nut out the pathway to chronicity I think the big problems we have to deal with is the presence of pathways and interactions. So when we have a great big dataset with lots of variables we can use to try and predict something, we have a problem trying to break it down into a few ‘things’ that predict our outcome using traditional stats techniques. That is because some things influence other things which in turn influence our outcome, but also some things may negate or inflate the influence of some other thing on our outcome (consider the effect of external stressors on a persons anxiety levels at a critical point in their pain history). We can’t possibly hypothesise all these complex relationships and model them under a traditional statistics framework; these frameworks work much better for confirming or refuting particular, relatively simple hypotheses. A paper that really gets me thinking about this is one Peter O’Sullivan has already referenced, Mazes, Conflict, and Paradox: Tools for Understanding Chronic Pain Cary A. Brown, Pain Practice, Volume 9, Issue 3, 2009 235–243. I think chronic back pain research is still in the exciting time of sorting out the key things that may be important for transition from acute to chronic, rather like Goldman did with his algorithm. Our clues are coming from quality data from a combination of study types, not only large epidemiological studies but also in depth qualitative and laboratory based studies.
[Reply]