Posts Tagged ‘surveys’

Disease labels in national surveys – the case of COPD

The new CDC Framework for COPD Prevention, much better thought of as a well developed agenda for applied public health, estimates that half of the people with COPD in the US have not been diagnosed.

The report – developed by a group of experts during a workshop in 2010 – proposes first among its four goals that the US improve the collection, analysis, dissemination, and reporting of COPD-related public health data. In particular, it highlights the need to develop and initiate new data collection within existing surveys, an obviously efficient and valuable objective. It recommends:

Refining the definition of COPD in existing surveillance systems by adding the terms ‘COPD,’ and ‘chronic obstructive pulmonary disease’ to the currently used ’emphysema’ and ‘chronic bronchitis’ terms.

This sentence brought me to a stop, frustrated that this change still needs to be made.

Our public health agencies must soon recognize some basic limitations of our surveillance systems: If they attempt to assess prevalence by asking people to report a physician diagnosis of some disease, they have no choice but to coevolve with diagnostic and popular nomenclature.

In this case, COPD has now been the dominant label for several years. Changes in awareness and labeling have been driven by many forces, including an improved understanding of the pathophysiology and natural history of the disease. But also by:

  • the rise of national organizations with names like the COPD Foundation, which was founded way back in 2004,
  • the marketing and education efforts of our own federal agencies (see the NHLBI page for example, or, well, the title of the report, which one downloads from the COPD page at CDC),
  • an increase in direct to consumer marketing of pharmaceuticals for the disease (such as this page for a branded formulation of tiotropium).

If we expect accurate and reliable prevalence estimates, we need to require that national surveys such as BRFSS, NHANES, and NHIS match the terminology in circulation. It is no longer acceptable to field surveys characterized by diverging epidemiological measures and popular labels.

A quick look at search frequency for COPD and emphysema (via Google Trends) shows that, in the US, search for COPD is more than twice as frequent compared to that for emphysema, and has been since about 2007. The dramatic, transient increases in search frequency for emphysema, which followed highly-publicized celebrity deaths attributed to emphysema (Johnny Carson in January of 2004) and mis-attributed to emphysema (Amy Winehouse in 2008), underscore the dynamic popular landscape of disease labels but also show the resilience of the longer-term trend.

Moreover, there are not large regional variations here; search for COPD (in blue) dominates that for emphysema (in red) across all regions of the country.

Inertia in the methods of our national surveys undermines the utility and value of their resulting estimates and, while it may make for more stable measures, assures that they are steadily providing information about an increasingly inappropriate category.

Hosted survey platforms need tools for panel studies

I use SurveyMonkey and a few other hosted survey platforms for a number of projects. Generally, these are great tools and I recommend them routinely to others.

One significant limitation is that they assume that the group of respondents completing a survey is unrelated to the group completing another survey.

In my work, this is often not true. Instead, it is more typical that we are surveying the same population on multiple occasions over time. For example, an individual may complete an entry survey and then complete additional ones at regular intervals during the project and upon exit.

In these cases we are, among other things, interested in change over time within an individual and among the group, and also in comparison to another group of individuals. Most often, we are examining the effect of some kind of intervention by studying change over time or the effect in an intervention compared to a control group.

However, since we can’t examine an individual’s responses over the course of the surveys, what we end up doing is exporting the discrete result sets from each and then importing them to a database where they are linked by some unique ID. It’s not straightforward but it’s entirely workable.

It would be very helpful to see one of the hosted survey platforms develop the ability to set up surveys appropriate for these kinds of panel studies. This would allow us to quickly review responses from a given person over time. At the group level, we could adjust aspects of the intervention on the fly without having to switch gears and set up the analytical environment.

In addition, it would be helpful to tie the responses of an individual in one survey to the questions or answers presented to the same individual in another survey. In other words, cross-survey logic. Right now, as far as I can tell, it’s missing from all the options I’ve used.