MR Partin, Minneapolis VAMC; M Malone, Minnesota Department of Health; M Winnett, Minnesota Department of Health; J Slater, Minnesota Department of Health; A Bar-Cohen, Minnesota Department of Health; L Caplan, Hebrew University

Objectives: To assess the effectiveness of two mailed interventions for increasing repeat mammography use among women 40-64 participating in a federally funded cancer screening program.

Methods: A total of 1558 women receiving mammograms through the Minnesota Breast and Cervical Cancer Control Program (MBCCCP) between June and December 1998 were randomized to three study groups: (1) mailed reminder about the availability of MBCCCP services two months before the rescreening due date (minimum intervention), (2) mailed thank you card within one month of the study qualifying mammogram, three mailed patient newsletters, and mailed program services reminder (maximum intervention), and (3) no mailings (control). The primary outcome was the proportion of women receiving a repeat mammogram within 15 months of their study qualifying mammogram. Outcome estimates based on medical records data and self-reported phone survey data were compared. Logistic regression models (adjusting for patient age, race, hysterectomy, month of qualifying mammogram and site characteristics) were used to estimate the association between treatment group assignment and these outcomes.

Results: The overall response rate for the survey was 81%. The model relying on survey data did not reveal any significant treatment effects (the minimum treatment group odds ratio was 1.10, with a 95% CI 0.81-1.5 and the maximum treatment group odds ratio was 1.01, with a 95% CI 0.76-1.36). However, the model relying on medical records based data DID reveal statistically significant treatment effects. The adjusted odds ratios for treatment effects in the model relying on the medical records-based data were 1.25 for the minimum intervention group (95% CI 0.96-1.62) and 1.35 for the maximum intervention group (95% CI 1.05-1.74). Efforts to explain the differences between the medical records and survey based results revealed a significant interaction between survey response and the medical records-based treatment effect. Among subjects who did complete the survey, there was no significant difference between treatment and control groups on the proportion rescreened through the MBCCCP within 15 months. Among subjects who did NOT complete the survey, there was a statistically significant difference of approximately 17.5% between the maximum intervention and control group on the proportion rescreened within 15 months (the maximum group having the higher proportion). Survey non-responders were significantly more likely to be non-white, unemployed, unmarried, to have less than a high school education, and to live in the 7 county metro area – the socioeconomic group which is the most likely to be eligible for MBCCCP services.

Conclusions: These findings indicate that the maximum intervention tested was very effective among the subgroup of individuals most likely to be eligible for the program services being promoted. However, if this study had relied solely on survey-based mammography information, our conclusion would be that this intervention was not at all successful.

Impact: These findings demonstrate that even a relatively low survey non-response rate can significantly threaten the validity of intervention study conclusions based on self-reported health behavior data, and imply that studies promoting interventions which target disadvantaged groups may need to rely on medical records or claims based outcome data to reach valid conclusions.