New research to help hospitals treat patients efficiently

Nicos Savva’s work aims to root out some of hospitals’ harmful profit incentives


In 30 Seconds

  • All over the world, hospitals are reimbursed for treating patients using a scheme that encourages cost efficiency. Along with the benefits of this prevailing reimbursement scheme, however, there is also an unhealthy tendency to select patients who are likely to bring in more income – a practice known as cherry picking.
  • Prof. Nicos Savva and coauthors have created a framework that uses game theory to reveal some unintended consequences that stem from efforts to reduce cherry picking and other profit-seeking practices. 
  • The researchers’ findings suggest that providing the right incentives for hospitals to treat patients effectively is complex and requires using data more effectively.

The latest research from London Business School’s Nicos Savva uses game theory to shed light on the dubious practices that stem from hospitals’ dominant model of reimbursement. 

Given existing financial incentives, many hospitals are encouraged to “cherry pick” profitable patients and to “upcode” patients with simple conditions in order to bump pay. At the same time, the market is divided into two types of hospitals: specialists that treat only the easiest cases and generalists that provide treatment to everyone. As this new research model indicates, cherry picking, upcoding and market bifurcation end up undermining our healthcare systems’ efficiency goals.

Consider this: Hospitals don’t compete based on prices. If you, as a patient, need your appendix removed, you typically wouldn’t pay out of pocket and, therefore, you have little incentive to comparison shop, as you might for a new oven.

So, how should hospitals be incentivized to keep costs down in the absence of direct competition? That trillion-dollar question has led to a working solution using indirect competition.

Here’s how the indirect competition works: Hospitals are not reimbursed based on the treatments and other services they provide to patients. Instead, they are reimbursed based on the patients’ diagnoses – or, more technically, the patients’ diagnosis-related group (DRG). The fee paid for each patient is equal to the average cost of providing care to patients who are assigned the same DRG across similar hospitals. It’s pretty ingenious: Hospitals end up with a profit if they can treat their patients for less than the fee – and, of course, the opposite is also true. If every hospital strives to keep costs below the system average, then the system average should be pushed down to an efficient level – similar to what’s seen when businesses compete based on prices.

“No one wants to be above average,” explains Professor Savva, who has researched hospital operations and healthcare management for years (and has an honorary appointment at Guy’s and St Thomas’ NHS Foundation trust, a large hospital system based in London). That is, no hospital wants to get stuck paying out for costs that exceed their peers’ costs when treating the same ailment. With this market mechanism in place, healthcare systems’ major payers – namely, governments and managed care organisations – keep their costs down, too. The system as a whole benefits. “Known as yardstick competition, this is such a clever system that has been adopted everywhere in the developed world,” Professor Savva notes. “It’s arguably one of the most impactful innovations in our academic field since World War II.” 

Discover fresh perspectives and research insights from LBS

“The game-theory model considers the hospitals’ evolving financial incentives”

How a headache begins

But no system is perfect. That is where recent research by Professor Savva – working with Laurens Debo and Robert A. Shumsky of Tuck Business School at Dartmouth – comes in. The coauthors have created a framework that uses game theory to shed light on some of the unintended consequences of the payment scheme over time in “Hospital Reimbursement in the Presence of Cherry Picking and Upcoding,” published in the journal Management Science. Many know that game theory was advanced by John Nash, the renowned American mathematician and protagonist of A Beautiful Mind. Game theory works out mathematically and logically the actions that players in a game should take to secure the best outcomes for themselves. What’s known as a “Nash equilibrium” in game theory occurs when players’ (or, here, hospitals’) strategies remain stable even when the other players’ strategies are known. In the paper, the game-theory model considers the hospitals’ evolving financial incentives as explained below.


Let’s start with a concrete example. After a hard knock on the head, a patient may be treated in a hospital for a concussion. Here, let’s assume that the hospital will be reimbursed by Medicare or federal health insurance in the United States. Back in 1983, when this scheme was first introduced, there were two DRGs for treating concussions at U.S. hospitals: one for concussions (code 32) and one for concussions with complications and comorbidities (code 31). Hospitals, on average, spent 33% more on code 31 patients, and so they were repaid 33% more, based on the average of all the relevant data. 


“When hospitals start dropping certain patients and cherry picking others, problems begin”

But within those two codes, naturally enough, concussed patients required more or less expensive care. Some complications required a lot more than code 31 paid out. At times, that was glaringly obvious well before a patient was admitted. Here, hospitals had an incentive to encourage such complicated patients to seek care somewhere else. 

When hospitals start dropping certain patients and cherry picking others, problems begin. Many times, it’s not that a hospital would flat-out refuse a patient who needed emergency treatment if they had numerous complications. It’s more likely a hospital would underinvest in equipment that would be necessary to treat the most complex cases (within DRGs) – and so they might be routed elsewhere. 

DRG iconDRG: A diagnosis-related group is a clinically meaningful classification of hospital patients based on similar resource use. It is used to determine hospital payments. In practice, one DRG may include patients that require more or less care.


Cherry iconCherry picking: Refers to the practice of selectively treating low-cost patients – that is to say, patients that will cost less to treat than what the hospital will be reimbursed (based on the DRG assigned). Cherry picking patients increases hospitals’ income.


Upcoding iconUpcoding: Intentionally assign patients to a more resource-intensive DRG than needed to increase income. In fact, some upcoding includes unnecessary treatments in order to justify receiving higher reimbursement.


Yardisck icon

Yardstick competition: A regulatory scheme for local monopolists (e.g., hospitals), where reimbursement is linked to performance relative to similar entities in the system. With its incentives to cut costs, this scheme serves as the theoretical underpinning for hospital reimbursement throughout the developed world.

Previous research has established that cherry picking happens. So, over the decades, as more data and better coding practices have become available, the number of DRGs in many reimbursement schemes has increased to help flatten out price differences within codes. Specifically, in the United States, Medicare now has three codes associated with concussions: 090, 089 and 088, depending on the existence and severity of complications and comorbidities – with payments increasing by 64% to treat the most resource-intensive DRG of the three. 

As Professor Savva’s research notes, in the United States, when DRGs were first implemented in 1983, there were 467 to choose from. Almost 40 years later, there were 761 – 63% more. A similar pattern has been seen in Europe, as the number of DRGs increased by 36% in Germany, 127% in the UK, and 239% in France between 2005 and 2011. 

What Professor Savva and coauthors have done is create a game-theory framework to shed light on some unintended consequences associated with recent efforts to reduce cherry picking and related profit-seeking practices with increases in the number of DRGs. 

Two paths to higher profits

The research indicates that having more DRGs works to reduce hospitals’ incentives to turn away complex patients, but with a caveat. That is to say, it works in some hospitals, but not others. At the same time, another surprising trend emerges.

The model shows that the market essentially splits into two groups: (1) Specialty hospitals that cherry-pick patients as much as possible and (2) generalists that treat all patients without engaging in cherry picking. In fact, the model shows that what individual hospitals do depends on what other hospitals are doing. For the former group, specialty hospitals (a growing subset of all hospitals these days), the source of profitability comes from avoiding the higher investment costs required to treat the very complex cases. For the latter group, generalist hospitals, the source of profitability comes from the fact that they treat more patients and are better able to recoup their higher investment costs. Equilibrium is found when the two types of hospitals coexist and are roughly equally profitable. From a system perspective, though, the equilibrium is inefficient because of cherry picking and underinvestment in cost-savings. 

Upping the ante with upcoding

A related unintended consequence of expanding the number of DRGs is that it increases the temptation to “upcode” – that is, assigning patients a DRG that will be repaid with a higher fee, even if it’s not clinically warranted. Returning to our original headache, concussions, remember DRG 088 receives a 64% higher reimbursement for income-seeking hospitals. As Professor Savva notes, “the thinner the difference between diagnosis codes, the more upcoding we see.”

The fact that upcoding happens has been documented in a variety of ways. In Germany, researchers found that when the DRGs for premature babies were based on the baby's birth weight, more babies were documented with lower weights when higher repayments were at stake. Another study along similar lines found that following a change in DRG prices, there was more evidence of upcoding where price jumps were higher. 

“With better administration, doctors can spend more time with patients, which is fundamental”

Upcoding, as a practice, works to undermine efforts to make the system more efficient when healthier patients are lumped in with sicker patients, reinforcing incentives for cherry picking. “Our concern is that cherry picking and upcoding can lead to bigger problems,” Professor Savva notes. “The most profitable hospitals in a system may, in some respects, be the most worrisome. We want to work to eliminate the wrong financial incentives out there.”

Using data well 

There has been a lot of talk about how data science, big data and machine learning can revolutionise healthcare. But Professor Savva is wary of hyped claims by tech evangelists promising too much. Working closely with doctors, hospital administrators and academics, he sees data science offering improvements, not revolutions in the near term. 

“I am optimistic that data science can make hospitals more productive,” he says. With better administration, doctors can spend more time with patients, which is fundamental.” (For more on this point, see: “The world is data-rich but analysis-poor.”)

For example, in their latest paper for Management Science, Professor Savva and coauthors suggest a way to realign incentives to encourage the efficient treatment of patients. They note that monitoring the total numbers of patients treated for each DRG – and then linking hospitals’ reimbursements to consider those numbers – could work to reward the hospitals that don’t upcode and punish those that do. 

Professor Savva’s studies of hospital management continue. He’s now working on a paper on the statistical estimates of hospital costs. After all, better data put to use can make our systems healthier.

Nicos Savva is Professor of Management Science and Operations at London Business School.