My Kingdom for an Algorithm

It was my turn to put on the academic teaching hat and present at journal club to my peers. The article I chose has been haunting me although I read it over 6 months ago. Isn’t it great when a scientific article, as opposed to some weird news story, gets under your skin and just won’t let go?

“Adjust to Target” appeared in print in the July 2006 issue of Diabetes Care. The study took place in several centers around the United States with a patient population of Type 2 diabetics using insulin. The basic premise of the study involved comparing two groups of study participants: the Simple Algorithm group was given a numerical algorithm by which they would adjust their insulin doses, and the Carbohydrate Counting group was given Carbohydrate ratios for determination of their insulin doses, the ratios of which were also adjusted based on a separate algorithm. Algorithms for everybody! At first glance, the algorithms appear dizzying. In fact when I first showed the page to one of our wonderful certified diabetes educators, she blinked and said, “They got patients to follow that?” My thought exactly, but later I realized the subjects were all in close contact with the study team, in fact weekly, for adjustment of the insulin doses. Overall the study results are quite impressive. Keeping in mind that they did have some not insignificant hypoglycemia occur during the study, the results are still enviable: an average drop in A1C to 7% or less by participants in both groups by the end of the study period of 24 weeks. In fact many participants achieved this goal A1C by 12 weeks. My dream, my dream as an Endo…

Despite my reluctant acceptance of the obvious fact that this study employed massive amounts of support staff to make calls to patients on a weekly basis, perform study visits are regular intervals, and a dedicated group of participants with little drop-out, I am haunted by this study. Why? Mainly because the results were so darn good, but also because they used algorithms for adjustment of the insulin doses; and it worked well.

Many of us in medicine look at algorithms with some degree of disdain. Especially among the internist “thinking” group, algorithms almost seem lazy to some degree. Why would you depend on an abstract group of numbers when you could (and SHOULD) be looking at the patient and knocking all possibilities around in your head to determine the right thing to do, all within 20 minutes of appointment time? Come on—that is what we internists DO!

Determination of insulin doses is not quite random of course. We do depend on carbohydrate counting as our gold standard along with some input from our calculated insulin sensitivity factors. But even those are determine based on the patient’s reported insulin usage, which can be wildly wrong for them in either direction (high or low), or a formulaic range of numbers used to decide the “average” dose for a patient of this size. On days when I have calculated the carbohydrate ratio, then the insulin sensivity, stop to review examples of calculating it with the patient, followed by numerous pointers of how to adjust if they are exercising/sick/menstruating/stressed/previously high or low/pregnant etc, I remember this article. And I wish for an easy algorithm to drop out of the sky. The Adjust to Treat study seems even more useful given that one of the algorithms was simply number-based instead of incorporating the carbohydrate counting which many of my patients cannot master for various reasons.

Today, after presenting the Adjust to Treat article, I spoke with one of our diabetes educators who was also very interested in possibly using it in clinic. Between us, we thought, perhaps we can do this in our outpatient clinic by alternating patients between my clinic and theirs. Twenty-four weeks may be a bit optimistic for a timeline, but a year seems reasonable.

After being in clinical practice for a few years after my training, I have already started realizing there are some battles that can be fought with different weapons, some seemingly less “academic” than others. Algorithms fall into the category for me. Not that developing them is not smart—the use of them is my issue. Some could argue that much of life is an algorithm of some sort, organized in our minds based on knowledge, teaching and our experiences. So why not use a mathematical algorithm for a medication? With everything in medicine being more challenging than they told us in medical school, who’s to say a little help from an unbiased algorithm isn’t just what we need to help our patients reach their goals (and get me some better sleep at night knowing it)? Maybe Adjust to Target also includes our methods, also known as “learning.”