Using Remote Tracking to Detect and Deter Medications Theft in Malawi

Studies of drug theft confront significant measurement difficulties that can impede appropriate policy designs: (1) identifying theft and distinguishing it from mismanagement or poor accounting is extremely difficult due to incentives for obfuscation; (2) identifying where in the supply chain theft occurs is not possible using most existing measurement protocols; and (3) the precise mechanisms linking policy interventions to outcomes are difficult to determine. We have designed our project to resolve these challenges. Our intervention involves two treatment arms:

1) The first treatment arm is a top-down government-led intervention involving the provision of information to clinic officials via posters. The information in the posters will convey that technology is being used to monitor drug deliveries in that specific clinic and the penalties for being discovered with stolen drugs.

2) The second intervention is a bottom-up community-led intervention that will improve the capacity of communities to monitor and hold officials accountable for drug theft. It will involve workshops with Health Centre Advisory Committees (HCACs) providing information regarding how to learn of drug delivery dates, drug availability, clinic responsibilities, and a reporting hotline. Committees will also receive a stipend for transport and communications costs for committee members to conduct site visits at the clinics.

2019-04-15

2019-05-15

Share of drug cartons per clinic delivery that fail to reach their destination.
Share of cartons stolen upstream or downstream from a clinic

We will use GPS and Bluetooth tracking devices to identify the point at which a drug leaves a warehouse, truck or clinic. Drug cartons that fail to reach a clinic (upstream theft) or disappear from a clinic after delivery (downstream theft) will be coded as having been stolen.

Patient health outcomes, including access to drugs, cost of drugs, perceptions and knowledge of corruption, and perceived efficacy of anti-corruption efforts (by both citizens and government).

We will investigate if medications theft affects community welfare outcomes – e.g., access to healthcare, the cost of healthcare, and health outcomes. One of our research questions is to evaluate if interventions designed to reduce theft improve these welfare outcomes or if it is the case that they perversely affect healthcare by increasing the price or limiting the supply of medications in surrounding markets (i.e., open-air markets or private pharmacies).

We will randomly sample clinics within the Southern region of Malawi from a pre-existing list produced by the Ministry of Health. Two hundred (200) clinics will receive tracking medications measurement at endline. Our goal is to identify at what point in the medications supply chain is theft most common: in transit to clinics; at clinics at time of delivery; at clinics while in inventory; or after dissemination to patients. We propose two interventions to evaluate how the government can reduce the level of theft.

These 200 clinics will also receive baseline and endline surveys of government officials, Health Centre Advisory Committee members, and citizens in the area surrounding the clinic. At these same 200 clinics, we will collect baseline, midline, and endline data on the cost and availability of medications in private pharmacies and markets surrounding the clinics to investigate the effects of monitoring on welfare outcomes.

To provide a refined measure of drug theft, we will affix GPS tracking devices to drugs at government warehouses, allowing us to map the procurement process from delivery to end users and to identify how our two interventions will impact the location, scale and rate of theft. We will construct an experimental sample of two hundred (200) clinics within the Southern region of the country, where levels of theft are higher.

To evaluate the effects of drug theft on patient health outcomes, we will survey citizens in the area surrounding these clinics using a random walk procedure to generate a sample size of 35 households per clinic area. Enumerators are either trained or will be trained in correct survey procedures. We will ask the household about their experience with health services over the previous three months. We will particularly focus on questions about the price of drugs, access to drugs, recent purchases of drugs, and experience of clinic services. Following the assignment of treatment, we will follow up with these same households with the same set of questions. Additionally, enumerators will visit local markets and clinics to assess the price and availability of drugs provided outside of the public sector at baseline, midline, and endline.

We will have three treatment groups and a pure control group. The first treatment group of clinics will receive information about the new monitoring system via official posters from the Ministry of Health. The second treatment group will receive the community intervention. We will hold workshops with the Health Facility Advisory Committee for the clinics in this group. Our trained enumeration team will provide them with information on ways to monitor and report medications theft from clinics. In doing so, we will build upon on pre-existing curricula developed by local civil society actors. Additionally, we will increase the capacity of communities to observe theft by providing them with information on the date and type of drug deliveries arriving in their local clinic, training on how to identify theft and hold clinic officials accountable, and a small stipend to cover their transport and communications costs. The third group will receive the information intervention at the clinics and the community intervention. A pure control group will not receive any interventions, and instead will experience only baseline and endline surveys and endline medications tracking with the GPS trackers.

Randomization done in office by a computer.

Clinics.

Yes

200 Clinics.

We will monitor 30% of the drug cartons for deliveries to 200 clinics. We anticipate this will involve tracking 2400 cartons. For the patient health outcomes, our proposed sample size is 7,000 subjects (200 clinics x 35 subjects per clinic).

50 clinics "top-down" treatment
50 clinics "bottom-up" treatment
50 clinics "top-down and bottom-up" treatment
50 clinics pure control (no intervention)

For the primary outcomes: We assume 200 clusters, 40 cartons per cluster and 30% of cartons being tracked. We also assume an overall population theft rate of 10%. At a power of 0.8, we have the power to identify treatment effects above 0.055 in a linear probability model with no covariates or blocking (conservative). At power of 0.9, we estimate a minimum detectable effect of 0.07. For the secondary outcomes: If we consider a power of 0.80, 200 clusters (with size 30), standard deviation of 0.4, the minimum detectable effect is 0.05. If we consider a power of 0.90, 200 clusters (with size 30), standard deviation of 0.4, the minimum detectable effect is 0.06.