Saturday, October 25, 2008

The Well-Tempered chart

This image above shows an example of automated data capture with the EMR; I am now tracking data on my visits for several different diagnoses. I can see that my annual check ups are stable, visits for common colds have seasonal patterns, and that I have changed my practice towards more chronic disease management and away from the treatment of acute, minor conditions.

The chart above is for the number of visits billed for hypertension; these are clearly declining. This is interesting to me, because it reflects the outcome of a number of changes I have made in my management of hypertension.

I use an automated BP machine in my practice; my staff do the BP readings, not me. I think the quality of BP readings in my office has improved as a result. Patients come in every 6 months if their BP is stable, consistent with current evidence. If BP reading is above goal, they are asked to drop by on a Friday (I am not in the office that day) to obtain an additional BP reading from the machine. These visits are not billed. I see the result remotely, and can send a message to obtain one more reading if needed, or to ask my secretary to book an appointment for medication optimization.

I also use home BP machines much more often. I have two loaner machines in my office.

The end result is good BP control, but fewer visits with the physician (see graph above). I don't think that routine BP measurement is a good use of my time, but I have switched from Fee for Service to a capitated payment system; this rewards efficiency instead of service intensity--there are pros and cons to that.

I can look at my data and plan further improvements because I now have "Well-Tempered charts". What I mean by that is that I have tried to enter good data, and to enter it consistently so that I can search it later. It was a learning process for me; I knew that my data would not be very good in the first year, and would then improve. I now enter "250" (diabetes) only when the patient is diabetic, and not when Impaired fasting glucose (or "pre-diabetes") is present. If I think the patient has angina, but I'm not sure, I will code the diagnosis as 785 (chest pain not yet diagnosed), comment "possible angina"; I code for angina, 413, only once I have the diagnosis. When I search my records, I now know that my diagnoses are highly specific (finding a code for diabetes means that the patient is truly diabetic). The searches may be less sensitive (may miss some diabetics), because some of my patients with pre-diabetes actually have the disease but have not been diagnosed yet.

This coding schema in my brain applies to important chronic conditions, because I really do want to identify patients with on-going problems that I want to manage better. I am less careful with minor conditions, such as colds; I may identify a cold as laryngitis or acute bronchitis (when they get an antibiotic).

Even though the EMR system allows me to use free text for conditions, I have limited this. I think trying to search for "DM II", "diabetes", "T2D" etc has far less value that coding the problem properly--even if you never misspell the condition.

Every prescription is entered in the EMR database; I avoid "free text" prescriptions whenever possible. Once I built my list of drug favourites, prescribing through the Multum database became much faster. All phone repeats are entered as prescriptions. I can now search through my prescriptions with a great deal of reliability.

Entering data is a pain; getting data out is the real gain. We need to think about the minimal requirements for the Well-Tempered chart, and I have outlined what I did above. I think coding correctly for about 10 common chronic conditions (DM, HT, depression, Asthma, COPD to start with) is a good start. Using the EMR to prescribe is helpful. Switching to labs that provide electronic results is a very good idea. None of this is terribly difficult, but it does take some effort to learn how to do it, and some discipline to enter data consistently.

JS Bach showed us that well thought out, orderly musical compositions are pleasing to the ear. I think we can learn from the Master, and apply his principles to the content of our records.



Ian Furst said...

Hey Michelle,

Interesting chart. I have to admit, I'm skeptical about the quality of data for two reasons. First is that the number of visits for HTN has dropped from 100 visits per month to roughy 20 in an 18month period. Even with a change in practice philosophy toward HTN that change is profound. Even more so if the patients are seen q6months. Second, the SD on the number of visits per month has gone up from roughly 10% to 100%. Both make me question whether this is a data entry error problem/change rather than true progress. Any chance of datamining appt instead of billing data? When did you change billing status?

Michelle Greiver said...

Ian, you are absolutely right. I should be much more skeptical about this data, instead of simply accepting what the computer spits out. Thank you for pointing this out.

I had another look at my data. We had been talking about the reports on our EMR on-line user group, and a company official stated that these pick up billings. The Dashboard report shows 101 HT entries for June 2007, and 62 for June 2008. I had a look at my billings with a code of hypertension (401) for June 2007 and June 2008: 41 patients billed in 2007, with 57 bills (some patients have multiple billing codes for each visit). 24 patients billed, with 43 bills in 2008. A clinical report for HT shows 52 HT codes entered in encounters for June 2007, and 35 in 2008--I sometimes enter multiple codes in each visit, but only the first code goes into the bill.

The reports on my Dashboard are clearly not picking up the data that I expect. I'm not sure what it is that they are picking up, and I'll ask. There does seem to be a decrease in visits and bills, although, as you point out, I am no longer sure of the data quality, and cannot make assumptions about degree of change.

This is a worthwhile reminder to always have a second look (and a third and fourth look) at your data.