Yi Yvonne Zhou, PhD; Terhilda Garrido, MPH; Homer L. Chin, MD; Andrew M. Wiesenthal, MD; and Louise L. Liang, MD
Outcome variables included annual adult primary care office visit rates, comprising appointments with physicians and physician extenders (nurse practitioners and physician assistants) in adult primary and urgent care (nonemergency) settings. Documented telephone contact rates included both scheduled telephone visits and unscheduled telephone calls to and from internal medicine and family practice physicians, nurse practitioners, and physician assistants. Statistical Analysis
We used the x2
test to look for differences in age/sex and the prevalence of diabetes and congestive heart failure between 5 groups of interest: the 323 296 adult members of KPNW, 18 094 KPNW members who had registered with KP HealthConnect™ as of September 2005, the 4686 cohort subjects, the 3201 subjects of the matched-control study, and the 3201 controls of the matched-control study. Differences in office visit and telephone call rates were symmetric about the mean, although not normally distributed based on the formal normality test. However, because percent changes in mean utilization rates are the most operationally meaningful way of examining trends, and parametric tests are robust to deviations from Gaussian distributions when samples are large,17
we used the paired t
test to assess the statistical significance of differences in utilization rates over time and across groups. As a matter of interest, the Wilcoxon rank sum test and the paired t
test yielded identical statistical significance.Cohort Study.
We calculated the difference in primary care office visit and documented telephone contact rates between the pre- and post-periods, assessing statistical significance with the paired t
We calculated the difference in primary care office visit and documented telephone contact rates in the pre- and post-periods for subjects and for controls, again assessing statistical significance with the paired t
test. In addition, we also used the paired t
test to assess the statistical significance of the variation in rate changes between the subject and control groups.
In general, KP HealthConnect™ Online users were older and included a higher proportion of members with diabetes than did the general adult membership. Similarly, cohort members were older and included a higher proportion of individuals with diabetes than did the larger user population. The differences in age/sex and the proportion of members with diabetes between the cohort and the general adult membership were significant (P
< .0001). By design, cohort subjects and both groups in the matched-control study did not differ to a statistically significant degree in terms of age and the proportion of members with diabetes. Table 2
summarizes these results. Annual Adult Primary Care Office Visit Rates
Baseline primary care office visit rates for cohort and matched-control subjects were slightly higher than the baseline rates for the entire region, consistent with the increased prevalence of chronic conditions. The baseline visit rate of the control group in the matched-control study was between the regional baseline rate and the subjects' rate. Cohort Study.
Annual adult primary care office visit rates decreased by 9.7%, a statistically significant decline from 2.47 to 2.24 office visits per member per year (P
< .001). Matched-control Study.
For the subjects in the matched-control study, the annual adult primary care office visit rate decreased by 10.3%, or 0.25 visits per member per year (P
< .001). The corresponding decrease for controls was 3.7%, or 0.08 visits (P
< .003). The difference between changes in primary care office visit rates for the 2 groups, 6.7%, was also statistically significant (P
< .003). Table 3
summarizes office visit utilization results for both studies. Primary Care Telephone Contact Rates
KPNW implemented new documentation procedures for telephone contacts during the study period; as a result, documented primary care telephone contact rates for the entire region increased by 24%. Documented primary care telephone rates for the cohort subjects increased by 15.6%. To evaluate the impact of access to KP HealthConnect™ Online in the context of this broad trend, we relied on the matched-control study.Matched-control Study.
The annual primary care telephone contact rate for subjects increased by 16.2%, or 0.32 documented telephone contacts per member per year, over the study period (P
< .001). The corresponding increase for controls was 29.9%, or 0.52 documented telephone contacts per member per year (P
< .001). The difference between these increases (13.7%) was also statistically significant (P
< .01), as displayed in Table 4
We evaluated the impact on office visit and telephone contact rates of patient access to an integrated multifunction electronic personal health record that included secure patient–physician electronic messaging. Annual adult primary care outpatient visit rates decreased by 6.7% to 9.7% for members using KP HealthConnect™ Online, and these members had a smaller increase in documented telephone contacts (16.2%) than the control group (29.9%).
Conducting a randomized controlled trial would have required fundamental changes to the KP HealthConnect™ system so that only patients randomized to the intervention were allowed to access online features. However, inconsistent member access to system features would have resulted in a prohibitive work-flow burden for physicians and healthcare teams. Additionally, random implementation of secure messaging would have confounded its true operational impact on office visit utilization. Nevertheless, our study controlled for individual patient factors, physician work styles, and regional trends as alternative explanations for reduced utilization.
Access to parts of the personal health record or other KP HealthConnect™ Online features may have influenced primary care office visit and documented telephone contact rates. However, an early evaluation of KP HealthConnect™
Online use among 1000 registered users found that more than 70% of sessions resulted in patient–physician messaging, indicating the importance and influence of this feature.
PDF is available on the last page.