We conducted a retrospective analysis of data generated from the randomized trial “An Internet Intervention to Promote Chlamydia Screening.” Study recruitment proceeded in two phases: phase I focused on the primary care office, and phase II targeted individual physicians from recruited offices. In the analysis of phase I, we examined factors associated with office recruitment. In the analysis of phase II, we examined factors associated with physician participation.
This study was approved by the Institutional Review Boards of the University of Alabama at Birmingham and the managed care organization.
Overview of Parent Study
The parent study, funded by the Agency for Healthcare Research and Quality as part of the second Translating Research into Practice initiative, tested an Internet intervention for primary care physicians and was performed in collaboration with a large, national managed care organization. The intervention was designed to increase Chlamydia screening of at-risk, young women by primary care physicians.
A series of Internet CME modules focusing on chlamydiosis were developed for the intervention. The goal of the instructional design of the online program was to create a multifaceted and multiphase online physician intervention based on current evidence of what is effective in CME. Delivery was via an asynchronous mode: physicians could log on at any time to participate. Email announcements and reminders with direct course links were used to alert physicians to the introduction of the course as well as three updates. Four separate modules were introduced quarterly. Components of the modules included (1) interactive unfolding cases with branching pathways designed to provide remediation based on the physician’s response to the case, (2) a quality improvement toolbox with resources to support office improvements in Chlamydia screening, (3) feedback to embedded questions so that participants could compare their responses to those of their peers, and (4) feedback of data on Chlamydia screening from the practice compared to peers within the overall group of practitioners. No online discussions were included. The intervention was designed and developed with Dreamweaver software (Dreamweaver MX, Macromedia, Inc., San Francisco, CA, USA) and used a SQL server database.
The Internet-based intervention for the control group was described to the participants as a CME course on women’s health for primary care physicians. Four modules, one each quarter, were offered to participants, and physicians could log on at any time to participate. The modules focused on women’s health issues unrelated to Chlamydia screening and included cardiovascular health and prevention of osteoporosis. The modules were text only and required participants to complete a post-test for CME credit. One category 1 CME credit was offered for each module. There was no mechanism for online discussions.
The intervention was designed for primary care physicians from internal medicine, family medicine/general practice, and pediatrics. Internists and pediatricians with a subspecialty board certification were not eligible. Physicians were randomized to an intervention or control group upon first logging on to the study Internet site. After one physician from a given office was randomized, all other physicians from the same office were assigned to the same study group.
Chlamydia screening rates, calculated by the managed care organization for each office, were based upon criteria from the Health Employers Data and Information Set (HEDIS) and provided the main outcomes for the parent study. The denominators for the rates were at-risk women identified from the 2001 HEDIS Technical Specifications applied to administrative data in the calendar year 2000. The HEDIS specifications were designed to identify women between the ages of 16 and 26 years who, based on health care services reflected in administrative data, were sexually active and therefore at risk for Chlamydia infections. HEDIS measures used pharmacy data (NDC codes) and claims/encounter data (ICD-9-CM and CPT-4 codes) to identify these health care services, which included pregnancy-related services, contraceptive prescriptions, screening for cervical cancer, and sexually transmitted diseases. The numerator was the number of women in the denominator who had claims data evidence of laboratory testing for Chlamydia during the baseline calendar year.
Recruitment and Enrollment
Phase I recruitment occurred at the office level. Each eligible primary care office had at least 20 patients aged 16 to 26 years who were at risk for chlamydiosis based on HEDIS criteria. In November 2001, all potentially eligible physicians (n = 4673) in eligible offices (n = 978) were faxed recruitment letters (Multimedia Appendix 1) inviting participation in the study. Recruitment letters were faxed twice. Initial nonresponders were then sent invitations by courier, but not if another physician from the same office had already been recruited. Letters described the project in general terms, as an Internet-based intervention to improve the care that physicians deliver to their female patients, but it did not indicate Chlamydia screening rates as the focus. Because the main purpose of phase I recruitment was to maximize the number of recruited offices, when at least one physician from an office agreed to participate, that office was labeled as “recruited” and no additional recruitment effort was made.
In phase II of recruitment, all physicians who provided their email address in phase I were invited to log on to the study Internet site. The intervention was initiated in February 2002 with an email broadcast to all recruited offices. Emails contained the website address, which connected directly with the module. Recruited physicians received email reminders (Multimedia Appendix 2) monthly and then weekly for a total of up to 33 reminders over a 45-week period until they logged on or asked to be dropped from the study. Only 2 physicians withdrew, asking not to receive additional emails. Emails for 18 physicians were returned because of invalid email addresses, and 3 email addresses did not belong to physicians. Intervention and control group physicians received email reminders according to the same protocol.
For all analyses, study variables were either (1) measured at the office level (patients/office and physicians/office), (2) measured at the patient level but available only at the office level (Chlamydia screening rates), or (3) measured at the physician level (physician age, gender, ethnicity, type of degree, country of graduation from medical school). Office characteristics were obtained from managed care organization administrative data. Chlamydia screening rates were calculated at the office level by the managed care organization based on HEDIS specifications. Other office characteristics were derived from managed care organization administrative data, and physician characteristics were derived from the American Medical Association’s physician master file.
The analysis for phase I examined factors associated with office recruitment (Tables 1 and 2) among all eligible offices (n = 978), and the analysis for phase II examined factors associated with physician participation (Tables 3 and 4) among all recruited physicians (n = 392). An office was labeled as recruited if at least one physician from that office was recruited. Physicians were defined as recruited if they provided an email address for subsequent contact. Participation was defined as having logged on to the study Internet site, regardless of how much of the material was completed. For phase I office-level analyses, the outcome was a dichotomous variable indicating whether the office had been recruited. For phase II physician-level analyses, the outcome was a dichotomous variable indicating whether the physician participated in the intervention. Independent variables included office and physician characteristics as described above.
Statistical significance was assessed using the chi-square statistic for categorical variables and the ANOVA for continuous variables for the bivariate analyses (Tables 1 and 3). Logistic regression was used for the phase I multivariable analysis (). For the phase II multivariable analysis, generalized estimation equations with a logit link accounted for the clustering of physicians within offices (). Because we were mainly interested in examining the independent contribution of covariates to either office recruitment or physician participation, we did not engage in covariate selection exercises to optimize the predictive power of the multivariable models. Instead, two models were constructed, each containing all important covariates.
Multivariable logistic model for primary care office recruitment among all eligible primary care offices (n = 821; c statistic = 0.622)
Multivariable logistic model for primary care physician participation among all recruited physicians (n = 324)