Archive for August, 2006

Exposing the Myths of Health Information Infrastructure

Sunday, August 27th, 2006

There is lots of discussion today in communities across the country about health information infrastructure. As people consider the issue, I thought it would be helpful to explore some of the myths and misconceptions about specific approaches and strategies that have been suggested to provide for the availability of complete patient records when and where needed.

Myth #1: The patient-carried record

One of the most popular and persistent myths of health information infrastructure is the patient-carried record. The idea is that if every person just carried their complete medical record, then it would be available for use whenever necessary. The record could be stored on a smart card, a USB drive, or some similar small and portable medium. Every site of care would have readers, and new information created at each visit would be written to the patient-carried record.

This idea is very appealing in its simplicity and low cost. On first glance, it appears to solve the problem, assuming that everyone could agree on the format of the stored records and obtain the needed hardware/software to read and write them (which would not necessarily be easy).

However, there are two serious flaws in this approach. First, what happens when the patient-carried record is lost, damaged, or destroyed? This can easily happen in a car accident, for example. Not only would the record itself be unavailable for the immediate need, but there would be no way to easily reconstruct it since there is no backup. To solve this latter problem, each person could have a second, backup record that they keep at home or in another “safe” location. However, that backup record would also not be accessible when needed for care UNLESS there was a backup location that could be reached electronically, i.e. via the Internet. However, if there is a backup of the patient-carried record available via a secure Internet portal, then why do you need a patient-carried version at all? The patient-carried record itself is the real backup in this case, and a relatively expensive one at that (compared to having a backup of all the records at the secure Internet portal). Furthermore, medical records available via a secure Internet portal would immediately be accessible from anywhere in the world without additional hardware and software, eliminating the need for everyone to have readers for the patient-carried version.

The second flaw in the patient-carried record approach is the problem of keeping it updated. This approach assumes that all medical information is generated when the patient (and the patient-carried record) are present — allowing the patient-carried record to receive the new information. But when x-rays are interpreted or blood test results are generated, the patient is rarely present. How would such information get to the patient-carried record? It might be argued that the next time any medical care is needed, the patient-carried record could be updated with this new information. But how would that information get to the next site of care (since we don’t necessarily know in advance where it might be)? Where would the new information be “held” until it can be downloaded to the patient-carried record? Would the new information be e-mailed to the patient? In that case, what if the patient forgets to do the update? Or doesn’t have e-mail or a computer? Clearly, it would be problematic to keep the patient-carried record up-to-date.

This is not to say that there is no role for patient-carried medical information. An up-to-date summary of problems, allergies, medications, and recent lab results could be very helpful IF patients would carry them. However, depending on this as a solution for delivering complete patient information when and where needed is not realistic.

Myth #2: Your medical record stored on your home computer

This idea is that everyone could just keep their complete medical records on their own home computer. After all, many people are already doing this with their financial information by integrating the data from multiple institutions. However, aside from the obvious problem that not everyone has a home computer, this approach does not work for your medical records. Unlike financial data, medical information may be urgently needed on a moment’s notice, and most likely the need will not be when you are at home with access to your computer. How would your doctor or hospital get access to the record in your home computer? Theoretically, you could leave your computer connected to the Internet and enabled for remote access. But then each person would need to implement and operate a highly secure portal to their computer to assure that there was no improper access and that viruses, worms, or hackers did not damage or destroy their medical records. In addition, each person would need to provide for backup power and telecommunications capability to ensure 24/7 availability, not to mention off-site backup of the information so it could be recovered in a disaster. Clearly, such efforts by individuals would be both unrealistic and prohibitively expensive. So this is not a viable solution.

This does not mean that having your medical records on your home computer is a bad idea — it could actually be very helpful. But your home computer is not a good place to have the copy of your medical records that is intended to be available for your care whenever and wherever needed.

Myth #3: “Google-like” retrieval of your medical records

Everyone is familiar with the impressive search capabilities of Google and other Internet search engines. With just a few keywords, they can rapidly find relevant information from (literally) billions of web pages. Why not use this capability to find your medical records — wherever they are located — and make them available for your care? (assuming they were all electronic and accessible via the Internet in a way that protected your privacy)

First of all, if this could be easily done, Google and others would already be doing it. The fact that they aren’t immediately tells you that there are fundamental problems. In my view, the most important problem is that Internet searching represents a type of information retrieval known as “non-deterministic”. In plain English, this means that the results of the search are never perfect — not all the items that should be found are actually found, and not every item displayed is one that is really relevant to the search. This is not a criticism of the search methods — they work really well — but is just inherent in the use of techniques for finding relevant documents.

In contrast, “deterministic” searching is what is done with computer databases. When someone searches their Accounts Receivable database to see which customers have balances over $500, the expectation is that the result will include every customer with such a balance and not any others. In this case, if the search did not work this way, we would say that there was an “error” and that the software was not working properly. When you search your Contacts file for “Mary Turner,” you expect to only find that name and you’d be puzzled if “John Tucker” also showed up in the results.

One reason it doesn’t work this way when using keywords to search for documents is that the “relevance” of a given document is itself not completely clear, and often depends on the context of the use of those key words (as well as human interpretation). For example, a search for “diabetes treatment” is highly likely to find a document with the phrase “… and this has nothing whatsoever to do with diabetes treatment” or “… this is in contrast to diabetes treatment, which is outside the scope of this discussion.” While these contain the phrase we are looking for, they are unlikely to be of interest.

Another reason document searching is challenging is that documents themselves are “free text” — not formatted into specific “fields” with known values. It is not easy for a computer to figure out the major topics of a 5000-word document (even people often find this difficult). Contrast this to a database where each item is in a “field” with a specific known format and meaning (e.g. phone number). When you know exactly where to find a specific piece of information and what it means, then a computer can easily retrieve it when asked. These two different search methods are also known as information vs. data retrieval.

So when searching for documents based on keywords, there is no absolutely reliable marker in each document that an algorithm can use to determine if that document is really relevant. The process is more like “pattern recognition” — trying to decide if the words in a document form a pattern that is consistent with what the query is requesting. In contrast, when searching the Accounts Receivable for balances over $500, it is easy to look specifically at the balance field and decide if it does or does not meet the “over $500” condition.

Getting back to medical records, many of which are also “free text,” it would clearly be unworkable to use document retrieval methods to find them. It would not be an acceptable response to a request for your records to locate 60-80% of them while also finding many records belonging to others. To be useful, a medical record retrieval method must find 100% of your records and none belonging to anyone else. Because of their inherently non-deterministic nature, no document retrieval method can do that.

But, you might ask, why not just label every one of my records with my name and a unique identifier so it can easily be found? That would solve the problem, but you’d no longer be using document retrieval (where you look for words), but database retrieval where you look in specific fields for specific values. The latter is not what Google and the other Internet search engines do.


Hopefully, the discussion of these myths will be helpful to you in considering how to approach the development of community health information infrastructure. For more information on a feasible and practical approach for building health information infrastructure in communities, please check out the previous posting on health record banking. As always, your feedback, comments, and additional thoughts are welcome.