Designing Incentives to Combat Urban Diabetes in India

The experimental sample will first be randomly divided into three main groups: Control, Incentives and Monitoring-only. The Incentives group will have their behavior monitored daily and receive incentives (in the form of cell phone recharges) based on their performance. The Monitoring-only arm will have daily monitoring but no incentives, which will allow us to isolate the effect of incentives.

Within the Monitoring-only and Incentives groups, individuals will receive pedometers that will measure the number of steps they walk every day, and will be given a step target of 10,000 steps per day. The Incentives arm for each of these groups will receive incentives based on whether or not they meet the step target. Both Monitoring-only and Incentives groups report their daily measurements using an automated toll-free phone system, and receive feedback by SMS based on whether they reach their target.

The structure of the incentive contract offered within the Incentives arm will be randomly assigned to one of three sub-treatments, meant to test time-preference: 0-lag delivery (daily payment), medium-lag delivery (weekly, i.e., participants receive payments once a week for that week's behavior), and long-lag delivery (every four weeks, or "monthly"). This design allows for identification of the effect of payment lags on behavior through both intra-arm comparisons (e.g., within the weekly delivery treatment, the lags change as patients approach the incentive delivery date) and cross-arm comparisons. To test the predictions about additive separability, the medium-lag subtreatment will be further divided into three subgroups: two with "threshold" contracts which require participants to meet goals on a threshold number of days in a week in order to receive payment, and one with an additively separable "nonthreshold" contract. To benchmark the impacts of incentive design with the impact of incentive amount, some of the participants with the weekly nonthreshold contract will be randomly assigned to receive smaller incentive amounts.

In order to benchmark the impact of incentives with a common intervention to encourage lifestyle modification among high-risk populations, we cross-randomize 10% of the experimental sample to receive informational SMS messages about the prevention, management, and potential consequences of NCDs.

2017-01-30

2017-12-31

Our primary outcome for measuring the health outcomes of incentives is Hba1c; secondary health outcomes include BMI, Blood pressure, a fitness assessment score, and a mental health score. Our primary outcome for evaluating the impact of the incentive design (i.e., for comparing incentive sub-treatments) will be exercise, measured as the percent of days on which the participant met his or her step target. Our secondary exercise outcomes include the average number of steps taken per day as measured by Fitbit records, and self-reported exercise.

Mental health outcomes will be constructed as an average score from a subset of the Rand 36-Item Short Form Survey (SF-36) for health assessment. The fitness assessment score will be calculated as the average of z-scores of two fitness tests (a walking test and a sit/stand test). Self-reported exercise is an average of z-scores for 4 self-reported exercise-habit questions

In this study, we propose to evaluate the efficacy of short run financial incentives designed to motivate diabetics and pre-diabetics to modify their lifestyles. Rewarding people for healthy behavior has been shown to markedly improve many types of health behaviors in both poor, but the use of incentives in combatting chronic disease has been limited to date. Whether such programs can effectively mitigate the particular challenges of urban diabetes remains an open question. The proposed research thus aims to address two research questions: First, can incentives cost-effectively promote lifestyle change among diabetics in urban India? Second, how can we optimize incentive design to account for diabetics' discount rates? We will answer these questions via a RCT evaluating several variants of a novel incentive program targeted to the needs of urban diabetics.

Randomization is stratified based on baseline Hba1c cut-off of 8%, and the answer to an initial survey question on "impatience". Ordered treatment assignment lists for each stratum cell are randomly generated in the office by a computer using Stata, and treatments are assigned to participants on rolling basis.

Individual

No

3100 diabetics and pre-diabetics

3100 diabetics and pre-diabetics

The following treatment group distribution is approximate, since enrollees will be assigned to treatments according to baseline strata in a pre-determined order:
Incentive Groups:
T1 - Weekly Nonthreshold: 28%
T2 - Weekly Threshold 1: 25%
T3 - Weekly Threshold 2: 10%
T4 - Monthly Nonthreshold: 5%
T5 - Daily Nonthreshold: 5%
T6 - Small Payment: 2%

Non-incentive Groups:
Monitoring Only: 6%
"Pure" Control: 19%

10% of the experimental sample will be cross-randomized into the "SMS" treatment

All power calculations use 5% statistical significance and 80% power. For comparing our incentives groups with the monitoring or control group to look at the impact on health: We have a minimum detectable effect (MDE), measured in standard deviations, of 0.12 SD for examining the effect of the pooled incentives treatments (weekly nonthreshold, weekly threshold, and daily nonthreshold) relative to control, and of 0.18 SD for the pooled incentives treatments relative to monitoring-only. The primary health outcome for these comparisons will be Hba1c. This calculation assumes a 5% attrition rate in endline Hba1c measurement, and a .5 correlation between baseline Hba1c and endline Hba1c. For evaluating the relative performance of our different incentive sub-treatments: As outlined in our pre-analysis plan, we will be testing whether the different incentive sub-treatments have different impacts on exercise. The primary outcome will be percent of days the participant met their step target. MDE’s for the primary comparisons expressed both in SD and in percentage terms (i.e., percent of days met step target), are as follows: -- Weekly non-threshold vs. Weekly threshold 1: 0.14 SD or .05 percent -- Weekly non-threshold vs. Daily non-threshold: 0.25 SD or .09 percent -- Weekly non-treshold vs. Monthly non-threshold: 0.25 SD or .09 percent

Pre-analysis Plan