Evaluating a preventive services index to adjust for healthy behaviors in observational studies of older adults

Analysis of outcome measures from nonrandomized, observational studies of people participating or not participating in health programs may be suspect because of selection bias. For example, fitness programs may preferentially enroll people who are already committed to healthy lifestyles, including use of preventive services. Some of our earlier studies have attempted to account for this potential bias by including an ad hoc preventive services index created from the patient's number of earlier clinical preventive services, to adjust for health-seeking behaviors. However, this index has not been validated. We formally evaluated the performance of this preventive services index by comparing it with its component parts and with an alternative index derived from principal component analysis by using the weighted sums of the principal components.

We used data from a cohort of 38,046 older adults. We used the following variables from the administrative database of a health maintenance organization to create this index: fecal occult blood test, flexible sigmoidoscopy, screening mammogram, prostate cancer screening, influenza vaccination, pneumococcal vaccination, and preventive care office visits.

The preventive services index was positively correlated with each of the following components: colon cancer screening (r = .752), screening mammogram (r = .559), prostate cancer screening (r = .592), influenza vaccination (r = .844), pneumococcal vaccination (r = .487), and preventive care office visits (r = .737). An alternative preventive services index, created by using principal component analysis, had similar performance.

A preventive services index created by using administrative data has good face validity and construct validity and can be used to partially adjust for selection bias in observational studies of cost and use outcomes.