Using Dynamic Incentives to Encourage Abstinence in a Drug Epidemic

Our primary intervention is a 12-week incentives intervention which provides participants with incentives for abstaining from drug use. Participants will be registered for a mobile app provided by DynamiCare Health and a provided with linked debit card. The app will prompt patients to submit saliva drug tests through their mobile phones on a random schedule (averaging three tests per week). For these randomly scheduled “contingent” tests, patients will receive immediate financial rewards in exchange for submitting drug-negative samples.

Within the intervention group, we will randomize six total incentive schedules: both a “High” and “Low” version of Escalating, De-escalating, and constant incentives. The schedules are designed so that both High schedules will pay similar amounts on average, as will both Low schedules.
• Escalating Groups, High and Low: In the Escalating groups, incentive amounts increase with every negative drug test up to a ceiling, and “reset” to the lowest amount when a test is positive or missed. Incentives in the High group will range from $4-$16, and $2-$8 in the Low group.
• De-escalating Groups, High and Low: In the De-escalating groups, incentive amounts increase with every positive or missed drug test (up to a ceiling), and decrease by the same increment with every negative drug test (down to a floor). Incentives in the High group will range from $10-$20, and $6-$12 in the Low group.
• Constant Groups, High and Low: In the constant groups

2021-09-29

2022-12-31

Our first primary outcome is the percent of follow-up tests that are negative, i.e., tests negative for opioids for people with opioid use disorder, or OUD; negative for cocaine for people with cocaine use disorder, or CoUD; and negative for opioids and cocaine for people with OUD and CoUD.
- To answer the research question “Do incentives increase abstinence?” we compare the control and incentives groups (pooled).
- To assess the most effective incentive (or incentive combination), we will implement the “smart pooling and pruning” method of Banerjee et al. 2021 among all treatment groups.


Our second primary outcome is the percent of contingent tests that are negative (where negative is defined as above).
- To answer the research question “Is high stakes or targeting more important for improving average compliance?” we will compare the escalating and de-escalating subgroups (pooled across high and low incentive levels).
- To answer the question “Do larger incentives matter?” we will compare the high and low incentive subgroups (pooled across escalating and de-escalating subgroups).

To answer the research question “Do high-stakes incentives come at the cost of increased variance of behavioral compliance?” we will again compare the escalating and de-escalating subgroups (pooled across high and low incentive levels). Our primary outcome will be:
- The variance of the individual-level proportions of contingent tests that are negative

To answer the research question “Do individuals respond to dynamic incentives?” we will compare the escalating, constant, and de-escalating subgroups (pooled across high and low incentive levels). This allows a clean test for whether there is an effect of dynamic incentives since all else is constant across these groups at the first test (including the payment amount). Our primary outcome will be:
- An indicator for whether the first contingent test is negative.

The primary form of missing data will be follow-up tests that are not taken. Our main approach will be to code these tests as “not negative” rather than “missing”. This lumps together missing and positive tests (a common approach in the literature). However, we will show the relationship between treatment assignment and test taking and utilize bounding exercises (e.g. Lee bounds) to clearly demonstrate the range of potential impacts missing data could be having on our results. This applies to all test results, both contingent and follow-up.

We will assess differences between the control group and incentive groups (pooled), as well as between the control group and the most effective incentive group (or incentive combination), as chosen by the “smart pooling and pruning method”, in the following secondary outcomes:

- Duration of time in treatment program
- Earnings
- Employment (dummy)
- Medical care utilization
o ED visits
o Inpatient stays
o Primary care utilization
- Total cost of care
- Incentive amount per negative follow-up test (i.e., “cost-effectiveness”)

We enroll adults adults with cocaine or opioid use disorders in urban Wisconsin. We begin with a one-week induction period, during which all patients will be registered for a mobile app provided by DynamiCare health and asked to complete an initial “non-contingent” saliva drug test. Patients who submit a first test will then be randomly assigned to one of our intervention groups, or to a control group that simply receives outpatient treatment as usual, for a 12-week intervention period. During this time, the intervention groups will receive their assigned intervention and the control group will receive treatment as usual. For all participants, the mobile app provided by DynamiCare Health will prompt remote “non-contingent follow-up” saliva drug tests every four weeks (including at the 12 week mark), which will be rewarded with $20 gift cards, to monitor abstinence behavior for study purposes. We will also conduct an endline survey with all participants at the end of the 12-week period.

Randomization was conducted by loading randomized allocation tables into the RedCap system

Individual

No

300

300

The sample will be divided between treatment arms in the following proportions:
1. Control 0.28
2. Escalating high 0.16
3. Escalating low 0.16
4. Deescalating high 0.16
5. Deescalating low 0.16
6. Constant high 0.04
7. Constant low 0.04