Sunday, January 5, 2014

Adjusting Insulin Dosage

A lot of diabetics leave insulin dosage to their doctors, and don't see their doctors very often. Others adjust their dosage themselves, cutting out the middleman.

Most drugs on the market, including the insulin Karl uses, have "patient information" and "physician information." We always read both.

As an engineer, I've been amazed by the primitive nature of the dose-adjustment information given to doctors. I realize that doctors are not as well-trained in basic numerical methods as engineers, but still...

By the way, one of the larger barriers to doing a good job has been whittled away recently. The newer Bayer Contour blood glucose meters, the Contour Next series, lets you enter the number of carbs and the insulin dose on the meter, which now has the full set of information (time, blood glucose, carbs, insulin) to let you analyze the results without typing pages of information from a hand-written log.

Anyway, like most Type 1 diabetics, Karl is on a long-acting insulin (Lantus) and a mealtime insulin (Novolog).

The numerical analysis challenge goes like this:


  • You want to give mealtime (bolus) insulin that's just right to handle the carbohydrates in the meal. This uses a carb factor, the number of grams of carbohydrate per unit of insulin. Get this right, and after the meal, blood sugar will be the same as before.
  • In addition, if blood sugar levels weren't perfect just prior to the meal, you want to adjust the insulin up or down to correct for this. This is the correction factor, the number of points the blood glucose value changes per unit of insulin, in the absence of food.
  • For long-acting (basal) insulin, you want a value that, in the absence of food, keeps the blood glucose levels steady all day.
Fair enough, but all three effects are going on all the time, so getting them all right is a bit of a trick.

I have not yet seen any publication where anyone even tries to do the number-crunching properly, or even to give useful rules of thumb, except for the long-acting insulin, where the rules are variants of, "Pay attention only to fasting (pre-breakfast) glucose levels, and adjust the dose of basal insulin, perhaps by one unit a day, until it yields numbers close to, but slightly less than 100, on average." 

Sure, they have rules of thumb for starting new patients on insulin, but that's about it.

Karl needed an increase in insulin dosages recently, as his numbers were creeping upwards, and what I did was as follows:

  • Adjusted the long-acting Lantus dosage upwards to get his morning numbers below 100, knowing that some of the problem was really that he also needed more mealtime insulin, and that I'd likely end up adjusting them downwards in the end, at least a little.
  • Looked at meals where Karl's pre-meal blood glucose levels were okay, so there was no correction, and thus the correction factor was not part of the picture. The difference between before-meal and after-meal blood glucose revealed the degree by which the carb factor needed adjustment. It turned out that he needed 9 carbs per unit instead of the previous 10.
  • Looked at meals where Karl's pre-meal blood glucose was high, requiring a significant correction, and what the post-meal blood glucose was. This gives a hint at what the correction factor should be. This required a significant correction, from 85 points per unit to 45 points per unit. I also double-checked this with a rule-of-thumb table about what the corrective dose is likely to be, based on daily insulin consumption, though of course I value actual results with Karl specifically to anyone's table about patient populations in general. Note that if the carb factor and correction factor are both out of whack, it's hard to adjust them both at the same time with any precision, but you can do them sequentially.
  • As predicted, the adjustments all called for an increase in insulin (not a surprise, since his blood glucose numbers had drifted upwards), and once the other two numbers were adjusted, the long-acting Lantus dosage proved a bit high, and we backed it off by one unit per day so far, and may back it off by one more.
This seemed to work pretty well. I probably ought to do regression analysis for more precision, and after we've switched over to the new Bayer Contour Next meters, I'm likely to take a stab at it.

The Bayer Glucofacts software that comes with the meter is pretty primitive, as I've said, but the better the control, the lower the standard deviation between readings, and the closer the average reading is to the target value.