Evaluating the representativeness of a cohort study of low back pain: Using electronic health record data to make direct comparisons of study participants with non-participants from the study population.

Representativeness is an important component of generalizability. Few studies have rigorously examined the representativeness of randomized trials or observational studies of pain or musculoskeletal conditions with regards to a wide range of factors beyond age, sex, race, and ethnicity. We conducted the first study of a pain condition that uses individual-level data to directly compare the enrolled study sample to the population from which it was drawn. We used electronic health record data from the Veterans Affairs health system to compare participants in an observational study of low back pain (n = 417) with the study population of potentially eligible non-participants who were contacted about the study (n = 15,218). There were no statistically significant differences between participants and non-participants for most factors examined, and differences when present were of small or very small magnitude. Participants were more likely to be older (odds ratio [OR]=1.02 per each additional year of age [95 % CI 1.01-1.03], p < 0.001), women (OR=1.59 [95 % CI 1.26-2.01], p < 0.001), have had a prior diagnosis of lumbosacral radicular syndrome (OR=1.37 [95 % CI 1.08-1.74], p = 0.01), and report lower pain intensity (OR=0.96 per NRS point [95 % CI 0.93-1.00], p = 0.04). Current smokers (OR=0.54 [95 % CI 0.39-0.75], p < 0.001) and people of Asian descent (OR=0.62 [95 % CI 0.39-0.98], p < 0.001) were less likely to participate. This study illustrates an approach to directly compare research participants with non-participants from the study population. This approach can be considered as a standard method to examine the representativeness of study samples in pain research. PERSPECTIVE: This article illustrates how electronic health record data can be used to directly compare the representativeness of participants in a study of pain to the study population from which participants were selected. This approach should be considered as a standard method to examine the representativeness of study samples during reporting.