After months of slumber the KBP blog has been awoken. The slumber was not a true slumber, the author has been otherwise engaged in a variety of activities keeping him (me) from writing. A series of events have been pushing me back toward posting again.
- APTA Education Leadership Conference presentation with my colleague Kelly Legacy called “Making Causal Inferences Explicit” – at some point soon I will voice over that talk and post (or just post the slides)
- An advance release of a paper about causal mechanisms was released to be published in PTJ – that will certainly be worthy of a post
- Engagement with the wonderful folks as APHPT and was just recently interviewed by their founder / president Mike Eisenhart. It was a great time and made me realize that some of this stuff I have been blogging about could be useful, so I may not want to throw in the towel just yet.
But, what has inspired me just this moment, is a PT In Motion article about an advance release of a paper coming out in the journal “Arthritis Care and Research” called: “Minimum performance on clinical tests of physical function to predict walking 6000 steps/day in knee osteoarthritis: An observational study“.
The take home message here will be that prediction without causation is wishful thinking. It is possible. Yes, it is true, data the establishes an association can then be used to predict. However, without causation the direction of the events are not known and our attempt to use prediction for intervention is left as wishful thinking. As an extreme example, if we demonstrated an association between a herniated disk and radicular pain then do we believe that removing the radicular pain through cryotherapy or medications or hypnosis makes the herniated disk go away? No. We believe that there is a causal association between the herniated disk and radicular pain, and while all of those mechanisms to intervene on radicular pain may have their place (I am neither recommending nor advocating), we are more comfortable that if the herniated disk was intervened on and went away that the radicular pain would go away (I do realize it is more complex than this, that there are other causes of radicular pain, etc).
To get back to the study at hand.
The purpose of this study was to identify minimum performance thresholds on clinical tests of physical function predictive to walk ≥6000 steps/day.
Methods: Using data from the Osteoarthritis Initiative, we quantified daily walking as average steps/day from an accelerometer (Actigraph GTM1) worn for >10 hours/day over one week. Physical function was quantified using three performance-based clinical tests: five times sit to stand test, walking speed (tested over 20 meters) and 400-meter walk test. To identify minimum performance thresholds for daily walking, we calculated physical function values corresponding to high specificity (80 to 95%) to predict walking ≥6000 steps/day.
Results: Among 1925 participants (age [mean±sd] 65.1±9.1 years, BMI 28.4±4.8 kg/m2, 55% female) with valid accelerometer data, 54.9% walked ≥6000 steps/day. High specificity thresholds of physical function for walking ≥6000 steps/day ranged from 11.4 to 14.0 sec on the five times sit to stand test, 1.13 to 1.26 meters/sec for walking speed, or 315 to 349 sec on the 400-meter walk test.
Conclusion: Not meeting these minimum performance thresholds on clinical tests of physical function may indicate inadequate physical ability to walk ≥6000 steps/day for people with knee OA. Rehabilitation may be indicated to address underlying impairments limiting physical function.
Emphasis has been added. It is an observational study. Data collected was cross-sectional. In the discussion about limitations the authors casually admit:
Third, the cross-sectional design allowed us to identify a relationship between physical function and physical activity, but not to draw conclusions about causation.
This is the third of four limitations which are mentioned. Of course, like a lot of studies, the limitations are mentioned but the implications of the limitations are not discussed.
In this study data was collected on performance (3 tests that are reasonable to expect to be associated with walking) and walking (as number of steps per day). Despite the data being collected at the same time (cross-sectionally) and the admitted limitations in drawing causal conclusions the study still moves ahead with the following causal assumption:
This seems perfectly reasonable. But so isn’t this:
The problem here is that this study did not explicitly test either of these assumptions, yet its conclusion is completely committed to the first assumption.
Not meeting these minimum performance thresholds on clinical tests of physical function may indicate inadequate physical ability to walk ≥6000 steps/day for people with knee OA. Rehabilitation may be indicated to address underlying impairments limiting physical function.
It is equally valid to have concluded:
Not walking ≥6000 steps/day may indicate these minimum performance thresholds on clinical tests of physical function for people with knee OA. Walking ≥6000 steps/day may be indicated to address minimum performance thresholds on clinical tests of physical function.
In fact, there is another hidden assumption. They have assumed that if someone does not walk ≥6000 steps/day that they cannot walk ≥6000 steps/day. This is a major assumption that has a huge impact on the conclusions drawn from this study.
In the end, this is a valuable contribution. It contributes to an understanding that can be built on with further work. But it does not allow prediction of the capability of walking ≥6000 steps/day. It maybe allows prediction of actually walking ≥6000 steps/day; BUT if we start with walking ≥6000 steps/day, it allows prediction of physical test performance. But these examples of prediction are based on associations. And predictions based on associations are less useful for clinical interventions.
But please keep in mind, that predication without causation is simply wishful thinking.