Predicting Out-of-Pocket Patient Expenses

The previous blog, Transparent and machine-readable prices (and surprise medical billing) was the second in a series on the Hospital Price Transparency Executive Order.

President Trump signed the Executive Order on Improving Price and Quality Transparency in American Healthcare to Put Patients First. CMS is currently requesting comments on its plan for implementing this order in this proposed rule. (Comment period ends 9/27/19)

Expected Costs

Section 3 (b) and later (c) is the beginning of the more investigatory components of the executive order. In Sec 3 (b) the President is clearly trying to shed some light on a very complex topic “expected out of pocket costs”. This is not a trivial problem. 

When getting any significant healthcare treatment at a hospital, your “expected expenses” are greatly determined by how otherwise healthy you are. A patient who is getting knee surgery who is perfectly healthy should expect one expense profile and a person who has diabetes, heart disease, as well as serious medication allergies, should expect a whole different set of expenses. 

In order to predict how much expenses might be, as well as how much of those expenses should be out-of-pocket, a lot depends on how much work the hospital needs to do to account for those individual differences in the health of a person. In some senses, the purpose of health insurance is supposed to average out those expenses over large groups- so that the “lucky and healthy” cover more of the costs for the “unlucky and therefore unhealthy”. 

Let’s ignore for a moment, that some people are unhealthy because of lifestyle choices and just focus on the economic reality of doing heart surgery on a person with emphysema or any surgery on a person who is morbidly obese. These differences in sickness are glossed over with terms like “complexity” that blur the differentials in both patient suffering and provider effort for these patients. “Complex”  procedures necessarily cost more money, and almost all data coding systems for procedures (DRG/CPT/HCPCS/etc) have mechanisms to account for these different levels of complexities and the corresponding efforts that go alongside them. 

But when a hospital chooses to bill a high-complexity DRG code for a surgical procedure, there are two ways that health insurance providers ensure the extra money was justified:

  • ICD diagnosis codes 
  • EHR chart reviews

Essentially, when a hospital says to an insurance company “John Doe’s surgery is going to cost 5x more than the average because he is that much sicker”, the insurance company always asks them to prove that. Sometimes this process is as simple as listing the other health problems that patients have, but when the price goes up, even more evidence is required. 

Figuring Out Out-of-Pocket Expenses

Essentially, figuring out the out-of-pocket expenses for a cash-pay or high-deductible patient needs to incorporate some kind of model of how sick the patient is, and that model is going to need to be trackable by patients. 

That is much harder than it seems. First, there is a language barrier, a patient does not think of their own medical history as “history of myocardial infarctions, complicated by atrial fibrillation” they think “I had a heart attack once and I get winded when I walk too fast sometimes”. There are ongoing efforts to ensure that we have good translations between medical language and natural language for medical conditions, but this work is relatively immature and has not focused on pricing transparency as a use case. 

Secondly, there is the issue of patient self-assessment of their own health status. Suppose an elderly man, Mr. Smith, is shopping for a knee surgery procedure and had a heart attack many years ago. He might think “well I had a heart attack, but then I lost 100 pounds and I took up running, so I should just shop under the ‘no complicating healthcare concerns’ section”. But every knee surgeon in the country is going to want to know that the patient had heart surgery, and as interoperability efforts improve, they will be able to garner this information much more reliably from previous medical records. This is going to result in Mr. Smith getting an upsettingly high bill because he is being charged for the actual risks that the provider was accounting for, rather than the risks Mr. Smith thought were valid. 

The heart of my point is not that “sometimes people will lie about the fact that they had a previous surgery” and definitely not that “it is more clinically complex to perform knee surgery with a victim of a previous heart attack”. Instead, my point is that patients and hospitals are not going to agree about how risky different patients are using the same mechanisms that insurance companies and hospitals previously did. 

So how can this be addressed? There needs to be some effort put into exposing the coding metrics that are currently used to justify procedure complexity into commonly understood check-box style price estimator tools. In the same way that the online portal of a moving company might say “How many beds do you have to move?” and “How many large boxes do you need?”, a “Procedure level calculator” for a procedure might say “Have you been diagnosed with diabetes?”.

Rather than “yet another form for the patient to fill out to solve a healthcare problem”, this is an opportunity for interoperable health records to help coordinate the pricing models correctly. A patient could use their primary physician’s EHR record to populate a hospital’s pricing app and get back “based on your medical record, we estimate that you are at mid-level pricing for this procedure”. 

But in order for those types of hospital or third-party apps to get the pricing right, there needs to be a consensus about how the patient’s healthcare record can be correctly interpreted into a pricing level. There are two things that could help with this.

Interpreting Records

First, having machine-readable information from the hospitals about what combinations of ICD codes justify which procedure levels would be helpful. Normally this information is maintained by payers, and hospitals are learning to meet those standards as they bill insurance companies for procedures. But in this case, these models need to be exposed from the hospitals to the patients, via machine-readable data: IF the desire is to figure out this problem on a case by case basis. 

The other option is to take a data-mining approach. The rules for which ICD codes get to be rewarded with higher-level HCPCS and DRG codes are very complex, but by studying large volumes of data in fee-for-service Medicare, Medicaid, and Medicare Advantage claims, it should be possible to create rough-granularity rubrics that could be understood by laypeople regarding what healthcare factors are going to impact price. 

So while the insurance rules which outline the justifications for higher-level hip surgery code might include 4 pages of industry jargon, the rules that would be derived from this study would be something like: 

If you have any 2 of the following problems, then you will need to be at the mid-level pricing tier for your hip surgery, and if you have 4 or more of them, you will be at the high-level price: 

  • Diabetes
  • A history of heart disease
  • Complex allergies to important medications
  • Etc

The reason why this is important to get right is that if there are objective ways for these costs to be estimated by patients in advance, then you can get healthcare providers who start to specialize and be price competitive at different levels of difficulty. In the same way that we seek different companies for service when we have one sink in our home that is broken than we do when we have four different bathrooms broken in a large factory, it is possible that the healthcare system will organize to compete for different complexities of patients. 

We know from our data analysis that this is already starting to happen with cancer treatment, for instance. If a person has stage 4 cancer, there are only a few places in the country where that person should probably go for treatment. But if you have only stage 1 cancer, you might be better off not going to one of those centers of cancer excellence. Being a high-value high-quality provider of both stage 1 cancer and stage 4 cancer might actually be an impossible optimization for a single organization. So there are substantial benefits to getting these issues right in pricing transparency efforts.

Fred Trotter

Fred shapes our software development and data gathering strategies, which doesn't stop him from getting elbow-deep in the code on a regular basis. He is co-author of the first Health IT O’Reilly book Hacking Healthcare, and co-creator of the DIRECT protocol mandated in Meaningful Use. Fred’s technical commentary and data journalism work has been featured in several online and print journals including Wired, Forbes, U.S. News, NPR, Government Health IT, and Modern Healthcare.