Premium
Statistical Projection of Clinical Subsample Estimates to a Survey Population
Author(s) -
LaVange Lisa M.,
Koch Gary G.
Publication year - 2007
Publication title -
journal of periodontology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.036
H-Index - 156
eISSN - 1943-3670
pISSN - 0022-3492
DOI - 10.1902/jop.2007.070110
Subject(s) - statistical inference , population , inference , projection (relational algebra) , causal inference , gold standard (test) , statistics , statistical model , variance (accounting) , econometrics , sample (material) , survey sampling , population projection , computer science , medicine , mathematics , artificial intelligence , environmental health , population growth , algorithm , chemistry , accounting , chromatography , business
Background: The goal of public health research often involves estimating clinical outcomes for a broad target population. The gold standard for the basis of such inference is a nationally representative survey that includes a clinical component. The cost of this gold standard can be prohibitive, and alternative approaches are needed. We propose a statistical methodology for projecting estimates from a clinical subsample to a larger survey population that has utility in periodontal research. Methods: The statistical methodology consists of fitting prediction models to clinical outcome variables that are available only for a subset of the survey population. Variables measured on the larger survey population are included in the model as predictors. The resulting prediction equations are applied to the entire survey population to produce estimates of prevalence for the clinical outcomes of interest. Methods for computing variance estimates for the model‐based predictions also are proposed. Results: This projection methodology was developed originally for use with a nationally representative sample of the veteran population in the United States as part of the National Vietnam Veterans Readjustment Study. Details of the projection methodology and illustration of its use are provided. Issues associated with the application of this methodology to periodontal research are discussed. Conclusions: The proposed projection methodology supports valid inference about clinical outcomes to a broad population using data from a more narrowly defined clinical subsample. This approach is useful in addressing important public health questions until such time as a nationally representative clinical study can be undertaken.