An Automatable Algorithm for In-Hospital Hypoglycemic Medication Dose Reduction Technology Development in Adults Living With Diabetes Mellitus

Recent estimates show that 27% or more of hospitalized adults are living with type 1 or type 2 diabetes mellitus (DM). ¹ The blood glucose (BG) target for nonacute care ward patients is: United Kingdom: 6.0 to 10.0 mmol/L ² ; United States: 5.6 to 10.0 mmol/L ³ ; Canada: 5.0 to 10.0 mmol/L. ⁴ Hypoglycemia is a very uncomfortable experience for patients, ² requires more nursing time for treatment, and is associated with longer hospital stays, and higher patient morbidity and mortality ^5,6.

A frequent preventable cause of in-hospital hypoglycemia is inadequate adjustment of DM medications when there is a “looming hypoglycemia”, defined as BG 4.0 to 6.0 mmol/L. ² An audit of four months of inpatient hypoglycemia events at our tertiary care hospital revealed an inadequate adjustment of DM medications in 64.3% of cases when there was already a BG below 5 mmol/L within the prior day.

The objective of this study was to develop and test an automatable algorithm to determine the most responsible hypoglycemic medication and recommend a decrease in its dose. This automatable algorithm could incite a new era of inpatient diabetes care technology development using artificial intelligence programming to integrate real-time hypoglycemic medication dose reductions into electronic health records. This algorithm could be written into software coding for integration into electronic health record systems in which the in-hospital patients’ medications and BG results are accessible in real-time.

Our study team developed an automatable algorithm that determines the most responsible medication for each in-hospital BG < 5.0 mmol/L event, and recommends the extent of its dose decrease. A BG < 5.0 mmol/L was chosen because it is below the desired glucose range for noncritically ill in-hospital patients.^2-4 The automatable algorithm (Table 1) contains three steps: (1) define the duration of action to determine the hypoglycemic medication(s) active in the patient; (2) establish a hierarchy of hypoglycemic medication potency; (3) recommend the dose decrease for the most responsible hypoglycemic medication.

OPEN IN VIEWER

Table 1.

Recommendation for Dose Decrease of the Most Responsible Hypoglycemic Medication: Follow Step 1, Step 2, and This Table for Step 3.

Most responsible hypoglycemic medication	BG 4.0-4.9 mmol/L (mM) a	BG 3.0-3.9 mmol/L (mM) a	BG < 3.0 mmol/L (mM) a
Basal insulin dose to be decreased is 5 units or less: b, c	20% dose decrease ________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 20% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 30% dose decrease. Bedtime to FBG drop > 6 mM: 40% dose decrease.	30% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 30% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 40% dose decrease. Bedtime to FBG drop > 6 mM: 50% dose decrease.	40% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 40% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 50% dose decrease. Bedtime to FBG drop > 6 mM: 60% dose decrease.
Basal insulin dose to be decreased is 6 to 29 units: b, c	10% dose decrease ________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 10% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 20% dose decrease. Bedtime to FBG drop > 6 mM: 30% dose decrease.	20% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 20% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 30% dose decrease. Bedtime to FBG drop > 6 mM: 40% dose decrease.	30% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 30% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 40% dose decrease. Bedtime to FBG drop > 6 mM: 50% dose decrease.
Basal insulin dose to be decreased is 30 units or MORE: b, c	5% dose decrease _______________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 5% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 15% dose decrease. Bedtime to FBG drop > 6 mM: 20% dose decrease.	15% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 15% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 20% dose decrease. Bedtime to FBG drop > 6 mM: 25% dose decrease.	20% dose decrease _________________ IF BG = FBG, then . . . Bedtime to fasting BG drop < 3 mM: 20% dose decrease. Bedtime to FBG drop 3.1 to 6 mM: 25% dose decrease. Bedtime to FBG drop > 6 mM: 30% dose decrease.
Premixed insulin or Regular insulin or Rapid insulin c	If dose to be decreased was 20 units or higher, then 5% dose decrease. If dose to be decreased was 6 to 19 units, then 10% dose decrease. If dose to be decreased was 5 units or less, then 20% dose decrease.	If dose to be decreased was 20 units or higher, then 15% dose decrease. If dose to be decreased was 6 to 19 units, then 20% dose decrease. If dose to be decreased was 5 units or less, then 30% dose decrease.	If dose to be decreased was 20 units or higher, then 20% dose decrease. If dose to be decreased was 6 to 19 units, then 30% dose decrease. If dose to be decreased was 5 units or less, then 40% dose decrease.
Correction scale insulin (Regular or Rapid) c, d	Decrease correction scale by 1 level of potency for the causative dose	Decrease correction scale by 2 levels of potency for the causative dose	Decrease correction scale by 3 levels of potency for the causative dose
Sulfonylurea e	Decrease by 1 dose increment	Decrease by 2 dose increments	Decrease by 3 dose increments

Step 1: To account for the potential delay in hospital pharmacy and nursing for any administered medication, we added 1 hour to the common duration of action for the following types of hypoglycemic medications: basal insulins 24 hours; neutral protamine Hagedorn (NPH) and premixed insulins 16 hours; regular insulin 8 hours; rapid insulins 6 hours; gliclazide modified release 24 hours; gliclazide immediate release 18 hours; glyburide 24 hours; repaglinide 5 hours. Step 2: To establish a hierarchy of hypoglycemic medication potency as the following from most to least potent: Rapid insulin fixed dose > Rapid insulin correction dose > Premixed insulin > Regular insulin (fixed dose or correction dose) > Basal insulin > Sulfonylurea. When more than one type of hypoglycemic medication was active at the time of the patient’s BG < 5.0 mmol/L event, the hierarchy of hypoglycemic medication potency determined the most responsible hypoglycemic medication. Step 3: To recommend the dose decrease for the most responsible hypoglycemic medication, which is dependent on the type and dose of this culprit medication that the patient had received.

a

More severe capillary blood glucose (BG) categories require a larger recommended dose decrease of the most responsible hypoglycemic medication. ^bFor basal insulin dose decrease recommendation, there are special suggestions if the hypoglycemia occurred in the fasting state. If the fasting blood glucose (FBG) was < 5.0 mmol/L, then the extent of recommended dose decrease of the basal insulin depends on the drop between the previous night’s bedtime BG to the FBG. ^cThe percentage dose decrease for insulins are to be calculated based on the insulin doses that were ordered for the patient. Dose decreases for insulin are to be rounded to the nearest integer unit. ^dThe correction scale insulin, also referred to as the sliding scale insulin, was determined to be the most responsible hypoglycemic medication if the patient did not have a fixed dose of insulin ordered, and it was the correction scale insulin received that was determined to be the most responsible hypoglycemic medication based on the algorithm’s step 1 and step 2. Correction scale insulin dose potency levels: none, low-dose, moderate-dose, and high-dose. ^eSulfonylurea dose increments: (1) Gliclazide modified release: increments of 30 mg per day (range 0-120 mg per day). (2) Gliclazide immediate release: increments of 40 mg per day (range 0-320 mg per day). (3) Glyburide increments of 2.5 mg per day (range 0-20 mg per day). (4) Repaglinide: 0.5 mg increments with each meal (range 0-4.0 mg per meal).

Our study trained certified diabetes educators (CDEs) to act reflexively like a software program running the automatable algorithm. From 23 October to 17 December 2023, the study’s CDEs worked shifts that covered 39% of the time. Alert messages were sent to the attending teams for 568 inpatient BG < 5.0 mmol/L events. Retrospective review of each of the algorithm’s recommendations showed that they were clinically-justified.

Compared to the same 8-week span in 2022, during this 2023 study, the number of BG < 4.0 mmol/L events decreased from 861 to 771 (P = .0002). The more severe BG < 3.0 mmol/L incidents decreased from 19.5% to 14.5% (P = .008).

Limitations of this algorithm include when a clinician chooses to decrease the doses of more than one of the hypoglycemic medications. Our algorithm is a potential starting point upon which future studies of DM technology software programming could build upon. Future studies could also integrate all classes of medications potentially influencing glucose levels, as well as other factors such as physiotherapy sessions, diet adjustments, and so on, into the software program that is connected with electronic medication ordering.