At Google, we believe that consumers should have convenient and secure access to all their health data so that they can be better informed and be more involved in their care. Recently, a data-savvy patient known as e-Patient Dave blogged about data that was imported into his Google Health Account from his hospital in Boston, Beth Israel Deaconess Medical Center. Once he saw his data in Google Health, he saw diagnoses that were both alarming and wrong. Where did they come from?

It turns out that they came from the billing codes and associated descriptions used by the hospital to bill the patient's insurance company. These descriptions, from the International Classification of Diseases (ICD-9), often do not accurately describe a patient because the right ICD-9 code may not exist. So the doctor or hospital administrator chooses something that is "close enough" for billing purposes. In other cases, the assigned code is precisely what the doctor is trying to rule out, and if the patient turns out not to have that often scary diagnosis, it is still associated with their record. Google Health faithfully displayed the data we received on Dave's behalf. We and Beth Israel knew that this type of administrative data has its limitations but felt that patients would find it a good starting point. Too often, this is wrong.

At Google, we are constantly learning important lessons from our users. Two days after we learned about this issue, I met with Beth Israel CIO John Halamka, the patient's physician Dr. Danny Sands, and e-Patient Dave himself. We agreed on a reasonable plan: Beth Israel will stop sending ICD-9 billing codes and will instead only send to Google Health the free text descriptions entered by doctors. Beth Israel is also working with the National Library of Medicine (NLM) to associate those free text descriptions with a more clinically useful coding system called SNOMED-CT, so that we can offer patients useful services like automatic drug interaction checking. The result will be more accurate and useful information in patients' Google Health profiles.

This week, all four of us were also at a conference called Health 2.0 in Boston. Dave's story, and the lessons we all learned, were the focus of much discussion. We are grateful to Dave for his openness and passion for making things right. We're also glad this happened because we and many others now better understand the limitations of certain types of health data and we are working with partners to improve the quality of the data before it gets to Google Health and our users. We look forward to sharing what we learn with the broader community. We also learned that the patient community is surprisingly interested in understanding these data issues. Dave and his doctor Danny Sands collaborated on an informative post about different data vocabularies used in different aspects of healthcare. The patient-controlled "data liquidity" that Google Health supports is clearly an important part of the future of health care. We are more committed than ever to putting consumers in charge of their own health information.