The third post in my series on Evidence Based Practice versus Patient Centred Care addresses the notion that the best source of evidence for patient centred care comes from the patient. I recall that when I was a speech-language pathology student in the 1970s, my professors were fond of telling us that we needed to treat each patient as a “natural experiment”. I was reminded of this recently when a controversy blew up in Canada about a study on Quebec’s universal daycare subsidy and the author of the study described the introduction of the subsidy as a “natural experiment” and then this same economist went on to show himself completely confused about the nature of experiments! So, if you will forgive me, I am going to take a little detour through this study about daycare before coming back to the topic of speech therapy with the goal of demonstrating why your own clients are not always the best source of evidence about whether your interventions are working or not, as counter-intuitive as this may seem.

Quebec introduced a universal daycare program in 1997 and a group of economists have published a few evaluations using data from the National Longitudinal Study of Children and Youth (NLSCY), one looking at anxiety in younger kids  and the more recent one describing crime rates when the kids were older . The studies are rather bizarre in that children who access daycare (or not) do not provide data for these studies – rather province wide estimates of variables such as likelihood of using daycare and childhood anxiety are obtained from the NLSCY which is a survey of 2000 children from across Canada, obtained biannually but followed longitudinally; then they estimated province wide youth criminal activity from a completely different survey rather than using the self-report measures from the NLSCY. Differences in these estimates (see post-script) from pre-daycare cohorts to post-daycare cohorts are compared for Quebec versus the ROC (rest of Canada, which does not have any form of universal childcare program). One author described the outcome this way: “looking at kids in their teens, we find indicators of health and life satisfaction got worse, along with teens being in more trouble with the law.” The statistical analysis and design are so convoluted I was actually hoodwinked into thinking youth crime was going up in Quebec, when in fact youth crime was actually declining, just not as fast as in the ROC. In actual fact, youth crime legislation and practices vary so dramatically across provinces, and particularly between Quebec and the ROC that it is difficult indeed to compare rates of youth crime using the variable cited in the NBER paper (rates of accused or convicted youths; for discussion see Sprott). Then they attribute this so-called rise but actual decline in crime to “the effects of a sizeable negative shock to non-cognitive skills due to the introduction of universal child care in Quebec”. Notwithstanding this nonsense summary of the results of these really weird studies, the most inaccurate thing that Milligan said is that this study was a “natural experiment” which is “akin to a full randomized experiment such as Perry Preschool, but on a larger scale”. But the thing is, a “natural experiment” is not an experiment at all because when the experiment is natural, you cannot determine the cause of the events that you are observing (although when you have enough high quality pairs of data points you can sometimes make decent inferences, NOT the case in this particular study). The economists know how to observe and describe naturally occurring events. They can estimate an increase in daycare use and changing rates of child anxiety and youth crime convictions in Quebec vs the ROC and compare changing rates of things between these jurisdictions. What they cannot do is determine why daycare use changed or reported anxiety changed or convictions for youth crime changed. To answer the question “why”, you need an experiment. What’s more, experiments can only answer part of the “why” question.

So let’s return to the topic of speech therapy. We conduct small scale randomized control trials in my lab precisely because we want to answer the “why” question. We describe changes in children’s behavior over time but we also want to know whether one or more of our interventions were responsible for any part of that change. In our most recently published RCT we found that even children who did not receive treatment for phonological awareness improved in this skill, but children who received two of our experimental interventions improved significantly more. Control group children did not change at all for articulation accuracy whereas experimental group children did improve significantly. In scatterpots posted on my blog, we also showed that there are individual differences among children in the amount of change that occurs within the control group that did not experience the experimental treatments and within the experimental groups.  Therefore, we know that there are multiple influences on child improvement in phonological awareness and articulation accuracy, but our experimental treatments account for the greater improvement in the experimental groups relative to the control group. We can be sure of this because of the random assignment of children to treatments which controls for history and maturation effects and other potential threats to the internal validity of our study. How do we apply this information as speech-language pathologists when we are treating children, one at a time.

When a parent brings a child for speech therapy it is like a “natural experiment”. The parent and maybe the child are concerned about the child’s speech intelligibility and social functioning. The parent and the child are motivated to change. Coming to speech therapy is only one of the changes that they make and given long waits for the service it is probably the last in a series of changes that the family makes to help the child. Mum might change her work schedule, move the child to a new daycare, enlist the help of the grandparent, enroll the child in drama classes, read articles on the internet, join a support group, begin asking her child to repeat incorrect words, check out alliteration books from the library and so on. Most importantly, the child gets older. Then he starts speech therapy and you put your shiny new kit for nonspeech oral motor exercises to use. Noticing that the child’s rate of progress picks up remarkably relative to the six month period preceding the diagnostic assessment, you believe that this new (for you) treatment approach “works”.

What are the chances? It helps to keep in mind that a “natural experiment” is not an experiment at all. You are in the same position as the economists who observed historical change in Quebec and then tried to make causal inferences. One thing they did was return to the randomized control trial literature, ironically citing the Perry Preschool Project which proved that a high quality preschool program reduced criminality in high risk participants. On the other hand, most RCTs find no link between daycare attendance and criminal behavior at all. So their chain of causal inferences seems particularly unwise. In the clinical case you know that the child is changing, maybe even faster than a lot of your other clients. You don’t know which variable is responsible for the change. But you can guess by looking at the literature. Are there randomized controlled trials indicating that your treatment procedures cause greater change relative to a no-treatment or usual care control group? If so, you have reason for optimism. If not, as in the case of nonspeech oral motor exercises, you are being tricked by maturation effects and history effects. If you have been tricked in this way you shouldn’t feel bad because I know some researchers who have mistaken history and maturation effects for a treatment effect. We should all try to avoid this error however if we are to improve outcomes for people with communication difficulties.

*******************************************

PS If you are interested in the difference-in-difference research method, here is a beautiful youtube video about this design, used to assess handing out bicycles to improve school attendance by girls in India. In this case the design includes three  differences (difference-in-difference-in-difference design) and the implementation is higher quality all round compared to the daycare study that I described. Nonetheless, even here, a randomized control trial would be more convincing.