Let me speak as someone who has published quite a bit of research using information derived from EHR data, most of which was ICD codes.
EHR data is fucking terrible. The quality is laughably bad. Despite there being standards for when specific codes should and should not be used, there is no way to actually enforce correct usage. This means coding practices may vary dramatically depending on exactly who is inputting the code. ICD codes get used incorrectly a fairly alarming proportion of the time. The most common deficiency is just missing codes. For whatever reason a lot of conditions just don't get recorded as ICD codes. You can look at the actual written doctors notes and see that they clearly noted that a patient had a specific condition but the ICD code associated with that often just never gets entered.
What's worse, its also common to have codes which are just flat out wrong. This can take many different forms but is usually the result of lazy medical history questionnaires.
"Do you have diabetes?" asks the questionnaire. Patient checks yes. Patient gets ICD-10 code E11. But E11 is the code for Type 2 diabetes. The patient has Type 1 diabetes, which is much less common. If I run a query to select all patients with Type 1 diabetes from my large EHR database I will miraculously find that more than 90% also show up in queries which select all Type 2 diabetes patients. Very interesting considering that the medically reported frequency of double diabetes in T1D patients is around 4%.
This is just one of many many many issues you see when working with EHR data. Findings derived from large scale EHR data can never be considered more than preliminary. Never.
I have personally had discussions with people who, even when getting their data from a single institution, have had to physically go around to different hospital departments and tell them to stop collecting certain types of data because they were so bad at it that they are making the rest of the data unusable.
Unless you have a carefully curated data collection process you will always be dealing with this issue. EHR has a lot of theoretical potential but the data quality issues negate almost all of it. It's really unfortunate.
Have you ever had to enter codes as a provider in Epic for instance? The UX is fucking terrible. Incentives are completely misaligned for accurate coding in academic institutions.
Finally the ontology of ICD codes are trash on their face.
Yet more of the laziest cynical AI generated garbage. AI winter can’t come soon enough.
Seriously I see nothing here of interest to a healthcare professional.
> The ICD-10-CM coding system is a remarkable catalogue of every imaginable medical condition, injury, disease, and even the seemingly improbable events that can befall a human being.
The thing with ICD is while it has an odd array of oddly specific and absurd codes it misses sufficient detail for some very basic common diagnoses.
> Our clinical vignettes serve multiple purposes. They provide a distinctive, engaging learning tool for medical students and healthcare professionals looking to familiarize themselves with the ICD-10-CM system.
Dive into the world of healthcare with ICDStories, a unique platform that brings to life the arcane ICD-10 medical coding system through intriguing and educative daily clinical vignettes. From commonplace conditions to the most peculiar occurrences, these stories not only demystify medical jargon but shed light on the vast spectrum of human health and ailment. Offering a fascinating blend of medicine, storytelling, and education, ICDStories is a must-visit resource for healthcare enthusiasts, professionals, and curious minds alike. Check it out at icdstories.substack.com!
Readit News
EHR data is fucking terrible. The quality is laughably bad. Despite there being standards for when specific codes should and should not be used, there is no way to actually enforce correct usage. This means coding practices may vary dramatically depending on exactly who is inputting the code. ICD codes get used incorrectly a fairly alarming proportion of the time. The most common deficiency is just missing codes. For whatever reason a lot of conditions just don't get recorded as ICD codes. You can look at the actual written doctors notes and see that they clearly noted that a patient had a specific condition but the ICD code associated with that often just never gets entered.
What's worse, its also common to have codes which are just flat out wrong. This can take many different forms but is usually the result of lazy medical history questionnaires.
"Do you have diabetes?" asks the questionnaire. Patient checks yes. Patient gets ICD-10 code E11. But E11 is the code for Type 2 diabetes. The patient has Type 1 diabetes, which is much less common. If I run a query to select all patients with Type 1 diabetes from my large EHR database I will miraculously find that more than 90% also show up in queries which select all Type 2 diabetes patients. Very interesting considering that the medically reported frequency of double diabetes in T1D patients is around 4%.
This is just one of many many many issues you see when working with EHR data. Findings derived from large scale EHR data can never be considered more than preliminary. Never.
I have personally had discussions with people who, even when getting their data from a single institution, have had to physically go around to different hospital departments and tell them to stop collecting certain types of data because they were so bad at it that they are making the rest of the data unusable.
Unless you have a carefully curated data collection process you will always be dealing with this issue. EHR has a lot of theoretical potential but the data quality issues negate almost all of it. It's really unfortunate.
Eh, ok.
Yet more of the laziest cynical AI generated garbage. AI winter can’t come soon enough.
Seriously I see nothing here of interest to a healthcare professional.
> The ICD-10-CM coding system is a remarkable catalogue of every imaginable medical condition, injury, disease, and even the seemingly improbable events that can befall a human being.
The thing with ICD is while it has an odd array of oddly specific and absurd codes it misses sufficient detail for some very basic common diagnoses.
> Our clinical vignettes serve multiple purposes. They provide a distinctive, engaging learning tool for medical students and healthcare professionals looking to familiarize themselves with the ICD-10-CM system.