TY - JOUR
T1 - Who wants a free brain scan? Assessing and correcting for recruitment biases in a population-based sMRI pilot study
AU - Ganguli, Mary
AU - Lee, Ching Wen
AU - Hughes, Tiffany
AU - Snitz, Beth E.
AU - Jakubcak, Jennifer
AU - Duara, Ranjan
AU - Chang, Chung Chou H.
N1 - Funding Information:
Authors Ganguli, Lee, Hughes, Jakubcak, Snitz, Chang, and Duara declare no conflicts of interest. Authors Ganguli, Lee, Hughes, Jakubcak, and Chang are supported by grant # R01 AG023651 from the National Institute on Aging. Dr. Ganguli is also supported by grant # K07 AG044395, Dr. Snitz is supported by grant # K23 AG08347.
Funding Information:
The work reported here was supported in part by grants # R01 AG023651, K07 AG044395, K23AG083479 from the National Institute on Aging, US Department of Health and Human Services.
Publisher Copyright:
© 2014, Springer Science+Business Media New York.
PY - 2015/6/27
Y1 - 2015/6/27
N2 - Neuroimaging research is usually conducted in volunteers who meet a priori selection criteria. Selection/volunteer bias is assumed but cannot be assessed. During an ongoing population-based cohort study of 1982 older adults, we asked 1702 active participants about their interest in undergoing a research brain scan. Compared with those not interested, the 915 potentially interested individuals were significantly younger, more likely to be male, better educated, generally healthier, and more likely to be cognitively intact and dementia-free. In 48 of the interested individuals, we conducted a previously reported pilot structural magnetic resonance imaging (sMRI) study modelling mild cognitive impairment (MCI) vs. normal cognition, and Clinical Dementia Rating (CDR) = 0.5 vs. CDR = 0, as a function of sMRI atrophy ratings. We now compare these 48 individuals (1) with all interested participants, to assess selection bias; (2) with all who had been asked about their interest, to assess volunteer bias; and (3) with the entire study cohort, to assess attrition bias from those who had dropped out before the question was asked. Using these data in propensity score models, we generated weights which we applied to logistic regression models reanalyzing the data from the pilot sMRI study. These weighted models adjusted, in turn, for selection bias, interest/volunteer bias, and attrition bias. They show fewer regions of interest to be associated with MCI/ CDR than were in the original unweighted models. When study participants are drawn from a well-characterized population, they can be compared with non-participants, and the information used to correct study results for potential bias and thus provide more generalizable estimates.
AB - Neuroimaging research is usually conducted in volunteers who meet a priori selection criteria. Selection/volunteer bias is assumed but cannot be assessed. During an ongoing population-based cohort study of 1982 older adults, we asked 1702 active participants about their interest in undergoing a research brain scan. Compared with those not interested, the 915 potentially interested individuals were significantly younger, more likely to be male, better educated, generally healthier, and more likely to be cognitively intact and dementia-free. In 48 of the interested individuals, we conducted a previously reported pilot structural magnetic resonance imaging (sMRI) study modelling mild cognitive impairment (MCI) vs. normal cognition, and Clinical Dementia Rating (CDR) = 0.5 vs. CDR = 0, as a function of sMRI atrophy ratings. We now compare these 48 individuals (1) with all interested participants, to assess selection bias; (2) with all who had been asked about their interest, to assess volunteer bias; and (3) with the entire study cohort, to assess attrition bias from those who had dropped out before the question was asked. Using these data in propensity score models, we generated weights which we applied to logistic regression models reanalyzing the data from the pilot sMRI study. These weighted models adjusted, in turn, for selection bias, interest/volunteer bias, and attrition bias. They show fewer regions of interest to be associated with MCI/ CDR than were in the original unweighted models. When study participants are drawn from a well-characterized population, they can be compared with non-participants, and the information used to correct study results for potential bias and thus provide more generalizable estimates.
KW - Epidemiology
KW - Propensity scores
KW - Standard error
KW - Survey sampling
UR - http://www.scopus.com/inward/record.url?scp=84929943597&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84929943597&partnerID=8YFLogxK
U2 - 10.1007/s11682-014-9297-9
DO - 10.1007/s11682-014-9297-9
M3 - Article
AN - SCOPUS:84929943597
VL - 9
SP - 204
EP - 212
JO - Brain Imaging and Behavior
JF - Brain Imaging and Behavior
SN - 1931-7557
IS - 2
ER -