Effectiveness and Efficiency of Government Health Insurance in India: Impacts of Strengthening Hospital Monitoring and Fraud Control

Treatment hospitals receive a government letter, signed by the CEO of the health insurance scheme, warning them that the insurance authorities are increasing fraud controls and that their audit probability has increased. The letters are intended to abruptly increase hospitals’ perceived threat of fraud detection, which has been shown to deter fraud in other contexts.
A second intervention is potentially scheduled for later. In this second intervention, the insurance trust would conduct hospital audits at randomly selected hospitals in both treatment and control groups. We are not yet sure this will be possible as it depends on the insurance trust's capacity. We will file an update to the registry entry to update this part of the protocol when plans have been finalized.

2024-03-28

2024-09-30

Our main outcomes will be: (a) composition of claims (including measures of upcoding, duplication, and unbundling of claims, estimated from analyses of claims and survey data); (b) claims volume, as observed in administrative data; and (c) average out-of-pocket charges paid by patients, as measured through exit phone surveys.

The key hypotheses are as follows:
1. If hospitals respond by reducing fraud, we should see a decrease in our measures of upcoding, duplication and unbundling.
2. If hospitals respond by turning away patients perceived as low-value from the standpoint of hospital’s profit (e.g., patients whose care costs would potentially surpass the pay rate absent upcoding or OOPP charges), we should see a decrease in (b).
3. If they compensate by increasing OOPP, we should see an increase in (c).
For all outcomes of interest, we will be estimating averages at the hospital-specialty level. The patient surveys have specialty-specific questions for a subset of specialties (in depth modules for pediatrics and obgyn, and lighter modules for cardiology, genitourinary and orthopedics) so analyses will vary in depth across these specialties. We will cluster the standard errors at the hospital level, since the randomization will be at the hospital level.
We will use a combination of administrative data, patient surveys, and government audits data.

Patient-reported care quality (including skimping on care that is less profitable in the absence of fraud),
Patient selection (turning away sicker/costlier or poorer patients).

Using historical data on claims, we identified private hospitals with relatively high claim volumes and values. Within these, hospitals were randomly selected for the "treatment" (letter intervention). We stratified the letter intervention randomization by what specialties the hospital offers (in particular whether or not the hospital offers the specialties targeted in the patient surveys), baseline average claims value, hospital share of district-level claims, and whether it is located outside AP state.

Public hospitals are not included in the RCT, but we will be able to use their claiming patterns over time as a useful benchmark since their incentives to manipulate claims are minimal.

Not available

randomization done in office by computer

hospital

No

340 hospitals

339 hospitals

169 Treatment, 170 Control