JMIR Publications

ABSTRACT

Background:

Telephonic outreach can be an effective but time and resource consuming method to encourage patient outcomes, and often resources are insufficient to reach all patients. Predictive modeling can help determine which patients respond best to outreach, in order to prioritize limited resources to be more effective at achieving patient health outcomes.

Objective:

Test effectiveness of telephonic outreach at different levels of patient likelihood of achieving outcomes, calculated by predictive models based on historical data.

Methods:

Within each of two rapid randomized tests in 2021 and 2022, a predictive model was created using historical data to calculate the likelihood of patients achieving a target health outcome by year end. For Study 1, the outcome was completion of an annual wellness visit. For Study 2, the outcome was reaching 80% percent of days covered (PDC) for a medication prescription. Participants included patients enrolled in a shared savings or risk contract for which the payer made monthly individual-level data available on target measures. Study 1: patients with no active MyChart account due for annual wellness visit between July 1, 2020 and January 31, 2021. Study 2: patients at with at least one unfilled prescription for a cholesterol, diabetes or hypertension medication between July 18 and December 19, 2022. Patients were randomized 1:1 to receive either no outreach (control) or telephonic outreach to encourage patients to complete health outcomes (intervention): Study 1: calls to complete annual wellness visit. Study 2: calls to fill medication prescriptions.

Results:

Study 1: Significantly more patients (n=26,861) completed their annual wellness visit in intervention than control overall (4.5% intervention vs. 0.3% control; RR=13.1; 95% CI: 9.7-17.7) and within each quartile of predicted risk. Study 2: Patient medications (n=6,784) had no significant difference in hitting PDC target between study arms overall (69.8% intervention vs. 68.6% control; RR=1.02, 95% CI: 0.98-1.06), but within the fourth quartile of predicted risk (patients most likely to hit target), intervention subjects were significantly more likely than control to hit their target (81.3% intervention vs. 76.3% control, RR=1.07; 95% CI: 1.01, 1.13).

Conclusions:

Precision medicine can help evaluate the effectiveness of patient outreach for different situations, allowing resources to be targeted to where they have most impact.