Response patterns in self-reported medical conditions: insights from the National Health and Nutrition Examination Survey (NHANES)

Abstract Background In the era of large-scale experiments, self-reported information has become a useful approach for gathering information longitudinally. The extent of subjectivity in self-reported medical conditions can vary considerably and can subject to a large degree of individual interpretation. Purpose To assess potential clustering among non-related self-reported conditions in the National Health and Nutrition Examination Survey (NHANES) over a period of 20 years (1999-2018). Methods We considered adult individuals (older than 20 years old) at the time of recruitment in all 10 cycles of the NHANES spanning from 1999 to 2018. For each individual, we gathered information on demographics, socioeconomic, examination data, laboratory variables and self-reported conditions included in the questionnaire instruments being available for all cycles. We focused only on questionnaire variables that are valid questions for both men and women. We examined potential patterns in clustering of responses among individuals in the same survey cycle and subsequently among individuals of different survey cycles. Results The study sample consisted of 55081 adult individuals. Female and male participants were equally represented in the study sample (52% and 48% respectively). The mean age of individuals at screening was 50 years (standard deviation 18). Participants were mostly non-Hispanic white (44%), followed by non-Hispanic black (21%) and Mexican American (17%). 41 self-reported medical conditions were available across the 10 cycles of NHANES and were considered in this analysis. The percentage of affirmative answers and missing information in the level of self-reported medical condition did not differ considerably across the 10 cycles for the vast majority of the medical conditions; but the percentage of missing information differed considerably across the different self-reported conditions and cycles ranging between 0% to 97% (Figure). Conclusions Surveys based on self-reported medical conditions may suffer from clusters and secular trends in provided responses or missing information among non-related conditions. Potential clustering may bias the data analysis and interpretation of the findings. The risk of clustering in individuals responses should be taken into account in analysis of self-reported conditions.Figure