Fusing Specialized Surveys of Rare Populations to Larger Surveys for Generalized Inference: Cross-Sectional Survey...

Area of Science:

• Epidemiology
• Survey Methodology
• Data Science

Background:

• Pharmacoepidemiology traditionally relies on broad surveys with limited behavioral detail or narrow, non-generalizable targeted studies.
• Existing methods present a trade-off between the breadth of drugs covered and the depth of behavioral data collected.

Purpose of the Study:

• To develop and validate a data fusion framework combining representative and specialized surveys for enhanced epidemiological surveillance.
• To apply this framework to estimate patterns of psychedelic drug use in the United States.

Main Methods:

• A cross-sectional survey study employing data fusion by combining a representative anchor survey with a specialized psychedelic drug survey.
• Calibration weighting was used to transport estimates from the enriched survey to the anchor survey population.
• Internal consistency (transport bias) and external validity (RMSE against benchmarks) were assessed.

Main Results:

• The fused survey design demonstrated good internal consistency with low average transport biases across demographics, health, and substance use.
• External validity was confirmed, with fused estimates aligning with probability-based benchmarks, showing improved RMSE for health and substance use.
• Recreational use of psilocybin, LSD, and MDMA was found to be significantly more common than medical use.

Conclusions:

• The "fused survey design" offers a rigorous and accessible method for surveilling rare behaviors, such as drug use, by integrating specialized data into broader surveys.
• This approach allows for generalizable inference on constructs not present in anchor surveys, expanding the capabilities of surveillance epidemiology.
• It provides a practical process for researchers to field enriched surveys with specialized questions without overwhelming respondents or exceeding survey constraints.