Whenever you go to the hospital (emergency room, doctor's appointment, or lengthy stay), the hospital needs to file claims with your insurance company so that they can get paid, and you can be covered. As in most situations where vagaries can cause major problems, certain systems have been created to ensure that communications are accurate. In the case of medical reimbursement, there are a number of different coding systems in place which map diagnoses and medicial procedures to a system of numbers and letters. This mapping is essential for the communications between hospitals and insurance companies so that the hospital, and the patient, can be correctly reimbursed and covered.
Back to topThe purpose of my project is to assist medical coders in making their claims as accurate as possible. To do this, I plan to use a knowledge base which will be updated with every claim put through the system. The goal of the knowledge base is to suggest codes that are likely to appear with the codes already entered, but which were left out. It is not uncommon for doctors to forget to write down everything they do, and this program will allow the coders to question the doctors concerning common procedures that they may have forgotten to take note of.
Back to topTo make sure that the suggestions offered by the program are meaningful, a series of threshold values will be put in place. These values would require a code be present with another code on X number of claims, or appear with another code at least X% of the time before being suggested to the coder. It will also be possible to use a combination of the two threshold types so that a code must appear with another code at least X times and at least X% of the time. This ensures that the codes being offered have a significant chance of actually being needed. In addition, the codes that are suggested will be updated as a user inputs more and more codes. Each new code will force the system to re-evaluate its suggestions to take into account all the codes that have been entered. This may cause some previously flagged codes to become unflagged, or some codes (especially complication codes) may become flagged as a result of the larger sum of codes entered.
Back to topTo facilitate my project and not bog it down with government rules and regulations inherent in real coding methods, I have devised my own coding system. I am using 40 unique codes - 10 diagnoses, 25 procedures, and 5 complications. All codes are 4 digit strings. For organizational purposes, diagnoses will follow the form 10xx, procedures will follow the form 20xx, and complications will follow the form 30xx. These choices have no bearing on the simplicity of the coding - they merely serve to keep the codes organized.
Diagnosis Codes
Procedure Codes
Complication Codes
Primary Diagnosis: 1003 - Broken Femur
Procedures:
2005 - Cast
2016 - X-Ray
2013 - Oral Drug B
Suggested Procedures:
Look for Complications:
3001 - Drug Interaction - 1% of claims, 3 total claims
3003 - Allergic Reaction - 10% of claims, 22 total claimsAs of now, I do not have the Swing requirement of the project complete. Having no prior experience has made the coding a little more difficult than anticipated. As far as suggesting codes, it seems that for claims with a large number (5+) of total codes, very few items get suggested, even though they may need to be. I am going to have to slightly rethink my strategy and instead of using all of the input, only using a max of 3-4 codes simultaneously to obtain recommendations. I could accomplish this by not throwing out suggested codes from higher in the decision tree as well, using all leaf nodes that did not ultimately appear in the claim. I have not implemented a persistent knowledge base yet either. Right now, the program only stores memories of the claims entered during the current session. Once the session quits, all knowledge is lost. I plan to fix this by writing out to file at the end of each claim, and loading the file at the beginning of each claim. Overall, I feel that this project could be a great asset to the way my company approaches its coding suggestions, as it becomes tailored to the user over time.
Back to top