Contact email

nkchoudhry@bwh.harvard.edu

Overview

  • This program was implemented across a large payor network and utilized predictive modeling to tailor insulin therapy adherence interventions in order to improve diabetes management. 
  • Three levels of progressively more targeted interventions were implemented to assess whether delivering more intensive insulin-adherence interventions to individuals with type 2 diabetes who were predicted to benefit most was more effective than delivering a lower-intensity intervention to an unselected larger group of individuals.

Department

Department of Medicine

Collaborators

Julie C. Lauffenburger, PharmD, PhD
Jennifer Lewey, MD, MPH
Saira Jan, MS, PharmD
Sagar Makanji, PharmD
Christina A. Ferro, PharmD
Alexis A. Krumme, MS, ScD
Jessica Lee, BA
Roya Ghazinouri, PT, DPT, MS
Nancy Haff, MD

Status/Stage of Development

Completed

Funding Sources

This project was supported through unrestricted funding from Sanofi, U.S.

Major Project Needs
  • None
Practice Setting

Health Insurance Network - Horizon Blue Cross Blue Shield of New Jersey

National/Policy Context

  • Poor adherence to diabetes therapy leads to adverse patient outcomes and increased costs.
    • Patients with diabetes requiring insulin therapy are an especially vulnerable subgroup due to unique therapeutic administration challenges, which include anxiety about self-injection, fear of hypoglycemia, and challenges re: the rising costs of insulin.
  • Most interventions to improve therapeutic adherence have been only modestly effective, in part because they are targeted to all eligible patients with diabetes rather than only to those who may benefit most.

Local/Organizational Context

  • We were particularly motivated to conduct this project because clinical pharmacists have been shown to be some of the most effective tools at improving adherence to medications. Therefore, we wanted to use predictive analytics to see whether directly pharmacist care to those predicted to be at greatest need improved outcomes at the population level. 

Patient Population Served and Payor Information

  • Horizon Blue Cross Blue Shield is the largest insurer network in the state of New Jersey. 

Leadership

  • Study staff members from the pharmacy-benefit management company regularly monitored the delivery of the intervention.

Project Research + Planning

  • Adherence: For each patient, the predicted risk of non-persistence to insulin was calculated by applying a regression-based algorithm to Horizon’s enrollment data as well as its pharmacy and medical claims data. The algorithm uses demographic and clinical information from enrollment data and claims as predictors. These predicted risks were then used to identify the patients that would be best suited for each type of intervention that was piloted during this program.

Tools or Products Developed

  • Text Message Program: The study investigators developed a text messaging program incorporated in a text messaging platform that focused on supporting medication-taking behaviors, lifestyle choices, and glycemic control. This text messaging program was included as one potential tool to optimize adherence to insulin. It was offered to some but not all participants based on the study arm and assignment to receiving the intervention based on predicted need.

Training

  • All pharmacists received training in medication therapy management and motivational interviewing.

Tech Involved
  • Telephone
Team Members Involved
  • Pharmacist
  • Physicians
Workflow Steps

  • Intervention Level
    • Participants received interventions based on their treatment arm. Arm 1 was non-targeted low intensity, Arm 2 was partially targeted moderate intensity, and Arm 3 was highly targeted high intensity. The type of outreach was similar in each arm, but the unselected 40% of patients assigned to Arm 2 did not receive any intervention, and the unselected 60% of patients assigned to Arm 3 did not receive any intervention. Thus, there was a lower volume of patients receiving outreach in Arm 2 and even fewer in Arm 3. 
    • After identification, all patients assigned to receive an intervention received:
      • A letter informing them about the pharmacist outreach
      • A reminder postcard
      • A small pillbox 
  • Initial Consultation
    • For all participants, the primary component of the intervention was an individually tailored telephone consultation conducted by a clinical pharmacist. 
      • Pharmacists attempted to reach everyone via phone to provide consultations.
      • Pharmacists engaged patients in discussions about individual beliefs, expectations, and barriers to treatments and provided counseling regarding strategies for achieving optimal glycemic control. 
  • Following the Initial Consultation 
    • Additional Resources
        • Quarterly mailings of patient education materials
        • Text messaging services
          • Participants in Arm 2 were offered enrollment in a weekly text messaging program focused on medication-taking behaviors, lifestyle choices, and achieving glycemic control. 
          • In Arm 3, participants were offered the text messaging program, with text messages delivered on a weekly, every 3 days, or daily basis. 
      • Follow-up Consultations 
        • In Arm 1, patients received up to 2 follow-up calls.
        • In Arm 2, patients received up to 6 follow-up calls. They also received 2 calls from their primary care clinician and/or pharmacists to clarify treatment issues or receive recommendations for therapeutic changes. 
        • In Arm 3, patients received up to 12 follow-up calls. They could also contact primary care clinicians and/or pharmacist as often as necessary.

Outcomes

  • All analyses of the effectiveness of the intervention were conducted at the population level (i.e., regardless of whether patients were assigned to receive intervention). 
  • Outcomes were evaluated using administrative claims data and linked laboratory data.
  • Demographics
    • 6000 participants (i.e., commercially-insured patients with type 2 diabetes using insulin)
  • Outreach
    • Telephone consultation percentage
      • Of those targeted, the following percentages of participants completed a telephone consultation with a pharmacist:  459 of 1861 patients (24.7%) in Arm 1, 342 of 1114 patients (30.7%) in Arm 2, and 251 of 731 patients (34.3%) in Arm 3.
    • Number of calls 
      • Patients with an initial consultation received the following mean (SD) number of calls: 1.7 (0.8) calls in Arm 1, 1.8 (1.1) calls in Arm 2,  2.1 (1.1) calls in Arm 3. 
    • Self-reported insulin adherence percentage
      • In the initial consultation, insulin adherence was self-reported by the following percentages of participants: 318 of 459 patients (69.3%) in Arm 1, 214 of 342 patients (62.5%) in Arm 2, and 148 of 251 patients (59.0%) in Arm 3. 
    • Text message plan opt-in percentage
      • Of those who were offered text messages, the following percentages of participants opted in to receive text messages: 85 patients (24.9%) in Arm 2; 13 patients (5.2%) chose to receive text messages daily, 7 patients (2.8%) chose every 3 days, and 31 patients (12.4%) chose weekly in Arm 3. 
  • Insulin persistence
    • Insulin persistence was the primary outcome and was measured in the 12 months after randomization to study arms.
    • As compared to the non targeted low-intensity intervention (Arm 1), neither the partially-targeted moderate intensity arm (Arm 2) nor the highly-targeted high intensity arm (Arm 3) significantly improved insulin persistence. 
    • In specific, insulin non-persistence did not differ significantly in Arm 2 (relative risk, 0.88; 95% CI, 0.75-1.03) or Arm 3 (relative risk, 0.91; 95% CI, 0.77-1.06) when compared with Arm 1. 
  • Glycemic control (changes in HbA1c level) 
    • Glycemic control was a secondary outcome and was measured by the mean change in HbA1c from baseline (prior to randomization) and the latest HbA1c value at the end of the 12-month period following randomization.
    • As compared to the non-targeted low-intensity intervention (Arm 1), the highly targeted high-intensity intervention (Arm 3) moderately improved levels of mean glycemic control. 
    • The partially targeted moderate-intensity intervention (Arm 2) did not significantly change HbA1c level.
    • In specific, when evaluated at the population level, glycemic control was best in Arm 3 (absolute HbA1c level difference, –0.25%; 95% CI, –0.43% to –0.06%). Glycemic control was similar in Arm 2 and Arm 1 (absolute HbA1c level difference, –0.15%; 95% CI, –0.34% to 0.05%).
  • Health care utilization/spending
    • The total amount of spending and number of office visits per patient did not differ amongst the three arms.
    • The moderate intervention (Arm 2) was associated with a small increase in hospitalizations. As compared to Arm 1, Arm 2 had more hospitalizations (odds ratio, 1.22; 95% CI, 1.06-1.41) and emergency department visits (odds ratio, 1.38; 95% CI, 1.24-1.53), which was an unexpected outcome of the study.

Future Outcomes

  • There are no specific plans at this time.

Benefits

  • The high-intensity intervention, as compared to the low-intensity intervention,  moderately improved glycemic control, a clinically relevant patient outcomes metric for diabetes control. These findings suggest that targeting patients for more intensive interventions based on both predicted risk of nonadherence and level of disease control may be more effective than untargeted approaches.

 

 

Intervention-Specific Challenges

  • Baseline persistence was fairly high in all 3 arms of this program. Thus, a ceiling effect may have limited potential benefits from interventions. 
  • Targeting a different patient population (i.e. one with a higher risk profile) might have increased the effectiveness of the interventions.
  • Outreach was performed by trained clinical pharmacists skilled at telephonic consultations, but these individuals were not a regular part of patients’ care teams, which could have reduced intervention effectiveness.
  • Administering interventions to patients who were predicted to be non-persistent without considering their actual glycemic control could have led to worsened outcomes requiring hospitalization.
  • Study Limitations
    • The acceptance rate within trial arms was slightly lower than in the initial power calculations. This may have been because rates were based on a slightly different population, and because this intervention specifically targeted patients who may have been more difficult to engage. 
    • Although predicted as nonadherent based on claims data, a large proportion of patients thought they were optimally adherent on self-reporting. This suggests further weakening of effectiveness of study measures and highlights the need for further optimization of predictive models used for randomization.
    • The persistence measurement was most likely not sufficiently sensitive. The 90th percentile threshold may not have detected enough variation in filling prescriptions. The threshold may also have detected differences only in very nonpersistent patients.

Glossary

  • Persistence was defined as not refilling their insulin prescription before a set threshold of time, which was assigned based on historical claims data from Horizon as the 90th percentile of time between each fill, adjusting for insulin type and quantity dispensed. 

Sources

  • Lauffenburger JC, Lewey J, Jan S, Makanji S, Ferro CA, Krumme AA, Lee J, Ghazinouri R, Haff N, Choudhry NK. Effectiveness of Targeted Insulin-Adherence Interventions for Glycemic Control Using Predictive Analytics Among Patients With Type 2 Diabetes: A Randomized Clinical Trial. JAMA Netw Open. 2019 Mar1;2(3):e190657. doi: 10.1001/jamanetworkopen.2019.0657. PubMed PMID: 30874782.

Innovators
  • Niteesh K. Choudhry, MD, PhD
Talk to the Innovators