The ongoing push for more descriptive medical diagnostic codes — the alphabet soup that is SNOMED CT, LOINC and, in the United States at least, ICD-10 — is motivated in part by a desire to improve data analytics for a host of clinical, billing and administrative purposes.
However, there can be an unfortunate consequence of such codes. The Journal of the American Medical Association has published the findings of a study suggesting that conclusions about reduced mortality rates among patients diagnosed with pneumonia may not have been entirely true.
The journal study found that, from 2003 to 2009, mortality rates did drop for patients with a primary diagnosis of pneumonia. Those with a primary diagnosis of sepsis or respiratory failure and a secondary diagnosis of pneumonia unfortunately did not share the same fate. When all three pneumonia diagnoses were combined, the mortality rate was “little changed” from the years prior to the study.
Here the impact of medical diagnostic codes can be readily seen. Focus specifically on codes referencing pneumonia and it appears that mortality is declining. Only when expanding the search to all relevant medical diagnostic codes referencing pneumonia does the bigger picture emerge.
This suggests that more descriptive medical diagnostic codes may make clinical data analytics more difficult. Compare the first research paper you ever wrote, which likely referred to only a handful of books from the elementary school library, to your graduate thesis, which sourced a veritable bookcase.
The issue isn’t the difficulty that descriptive medical diagnostic codes present so much as it is knowing what we can learn from them. One cannot expect an elementary school student to walk into the Boston Public Library and know where to find books about George Washington. Organizations are increasingly finding that it takes Ph.D informaticists, not to mention a clinical data warehouse, to get the most from that data.