Spillover

Two Common Blunders

1. Defining the Neighborhood
2. Equating Spillover with Physician-Spillover

Blunder 1 - Defining the Neighborhood

• Problem: Establish a juncture between the hospital and retail
• Solution: Define neighborhood of hospital
• Comment: But how?

Neighborhoods with different radii: 5, 10, 25 miles.

Popular Approach: Take the neighborhood to mean zip codes that surround the hospital or are located less than x miles away from the hospital (x=5, 10, 25). Use sensitivity analysis to alleviate doubt

Tri-State Problem: Montefiore Hospital (located in Bronx, NY 10467) and Lenox Hill Hospital (located in New York, NY 10021) draw patients from the same zip codes.

Popular Approach is Deeply Flawed!

1. In densely populated areas such as the tri-state where several hospitals are located next to each other, there is no absolutely rationale for assigning a zip code to hospital A as opposed to hospital B.

2. People do not go to the closest hospital. Some patients fly hundreds of miles to seek what they regard as best care, especially when confronted with life-threatening diseases. Having a relative on staff may certainly influence the decision.

3. People choose a hospital in part based on the reputation of the hospital (Cleveland Clinic for cardiology, Dana Farber for oncology, etc.) and the condition/disease that is afflicting the patient.

4. Sometimes, the choice of the hospital is dictated by the hospital affiliation of the treating physician (that hospital may not be the closest one to the patient).

5. The presence of nearby institutions (e.g., nursing homes, clinics, etc.) may severely skew the spillover neighborhood by altering the flow of patients to the hospital.

Although the popular approach has the advantage of being very simple and easy to implement, it suffers major flaws that severely compromise the findings it supports.

Blunder 2 - Focus on Affiliated Physicians

• Principle: Equate spillover to physician-spillover. Forget patient-spillover altogether.
• Implementation: Track only prescriptions of physicians affiliated with the hospital.

Even in the best hospital affiliation databases, a physician may be affiliated to an inordinate number of hospitals. More importantly, we don’t know if the physician ever set foot in the hospital or is 24/7 there!

Bridging Technique: Using Actual Data!

Our Approach

• Combine several facility-based patient data sources.
• That allows us to establish for each of the 6,700+ hospitals in the US and for each ICD-9 code (procedure or diagnosis) where the patients come from!
• In other words, we know perfectly the actual neighborhood of each hospital relative to the disease/condition in question!

Spillover Analysis Tool

For additional information about this product please send an e-mail to igor@bayser.com