Hypoglycemia in Hospitalized Hemodialysis Patients With Diabetes: An Observational Study

Abstract

Introduction:

Hypoglycemia and hyperglycemia affect outcomes in hospitalized patient. Patients with diabetes and end stage renal disease are prone to hypoglycemia and few studies have evaluated glucometrics to identify the incidence and risk factors for hypoglycemia in this population.

Methods:

We designed an observational retrospective review of 150 insulin requiring inpatients with diabetes receiving hemodialysis. We collected demographics, baseline characteristics, and glucometric data focusing on episodes of hypoglycemia with glucose cutoffs <70, <54, and <40 mg/dl. Detailed glucose and insulin data for 24 hours before and after hemodialysis was analyzed for each patient in context of a hypoglycemic episode. T-tests, one-way ANOVA, and chi-square tests were used for statistical analysis.

Results:

At least one glucose value less than 70 mg/dl was observed in 51% of hemodialysis patients, less than 54 mg/dl in 28%, and less than 40 mg/dl in 11%. Patients with hypoglycemia had a higher HbA1c, standard deviation of glucose (P = .0009) and higher total daily dose (TDD) of insulin by weight (0.34 units/kg vs 0.23 units/kg, P = .003). We observed a linear increasing risk for hypoglycemia with increasing TDD, with nearly 65% of hypoglycemic episodes occurring with TDD >0.20 units/kg. A majority (61%) of all hypoglycemic episodes occurred in the 24 hours prior to a hemodialysis session. Type 1 diabetes was independently associated with hypoglycemia.

Conclusions:

Hospitalized diabetes patients undergoing hemodialysis were found to have high rates of hypoglycemia. Our results support using a lower TDD of insulin in this population (<0.23 units/kg/day) and recommend special caution in those with type 1 diabetes.

Inpatient glycemic control has significant clinical impact on outcomes of hospitalized patients.^1-4 Hyperglycemia, hypoglycemia and glucose variability are each associated with poor patient outcomes.^5-7 The risk of hypoglycemia and mortality are particularly high among end-stage renal disease (ESRD) patients with one study reporting a 32% mortality rate in those requiring hospitalization for hypoglycemia.⁸ This increased risk is attributed to altered glucose homeostasis, decreased renal gluconeogenesis and altered insulin and drug clearance. In addition, malnutrition and autonomic neuropathy further attenuate the counter-regulatory, metabolic and hormonal responses to hypoglycemia.^8,9 Other complicating factors of uremia-induced insulin resistance and postdialysis insulin sensitivity added to the variability of diet, medication and timing of hemodialysis (HD) make glycemic control particularly challenging in the ESRD population.¹⁰

Several studies have implicated insulin dosing as a modifiable risk factor for hypoglycemia in general inpatients with and without renal impairment.^11-13 In a small outpatient study, episodes of hypoglycemia were noted to be more frequent in the post-HD period than pre-HD.¹⁴ To this effect, the American College of Physicians (ACP) recommends reducing insulin doses by 25% when GFR declines to 10-50 ml/min and further up to 50% when GFR goes below 10 ml/min.¹⁵ A suggested initial daily insulin dose of 0.2-0.3 units/kg has been recommended by a study evaluating hypoglycemia risk for inpatients with chronic kidney disease (CKD).¹⁶ However no evidence based recommendations specific for ESRD or for HD patients are yet available.

Hence, we performed this retrospective review of glucometrics in hospitalized ESRD patients with diabetes treated with insulin and receiving HD, to evaluate insulin dosing and other clinical factors associated with hypoglycemia.

Methods

In this observational study we included patients between 18 to 80 years with type 1 or type 2 diabetes mellitus, receiving insulin therapy and HD admitted to the University of Michigan Health System (UMHS) between June 2014 and September 2015. Patients with diabetic ketoacidosis (DKA), hyperglycemic hyperosmolar state (HHS), those on insulin drips or requiring corticosteroids (prednisone equivalent dose more than 10 mg/day) were excluded.

The primary outcome of interest was the number of patients with hypoglycemia, defined as capillary or plasma blood glucose less than 70 mg/dl. Secondary outcomes included patients with severe hypoglycemia (glucose < 40 mg/dl). Additional post hoc analysis of clinically important hypoglycemia (glucose < 54 mg/dl) as recently recommended by the International Hypoglycemia Study Group was also completed.¹⁷

Data collection was performed by electronic chart review. This included baseline demographic and medical information, baseline insulin requirements and hemoglobin A1C when available. Overall differences in baseline characteristics between patients with and without hypoglycemia were evaluated as was glucose variability, measured as the standard deviation (SD) of glucose values. All glucose values on the day of admission were excluded from analysis to remove bias of home treatment. Hypoglycemic episodes within a two-hour time frame were considered as those arising from incomplete treatment of the original episode.

We performed additional analysis on the relation of hypoglycemia to HD timing. The pre-hemodialysis day (pre-HD) was defined as the 24-hour period prior to the HD session and post-hemodialysis day (post-HD) was defined as 24 hours after the HD session. Early morning hypoglycemia was defined as a hypoglycemic episode between 3 a.m. and 8 a.m. To capture glucose values from at least one 24 hour pre-HD session and 24 hours post-HD session, only patients admitted for more than 3 days were included. Glucose values were collected on pre-HD days and post-HD days with at least 1 HD session and a maximum of 3 sessions per patient, to allow an adequate sample size for analysis.

This study was approved by the local institutional review board with a waiver for informed consent due to the retrospective design.

Statistical analysis: Comparisons were made between patients having hypoglycemia versus not having hypoglycemia using T-tests and one-way ANOVA for continuous variables and Chi-square tests (or Fisher’s exact test) for discrete variables. A P-value < .05 was considered significant. Logistic regression models to predict hypoglycemia were constructed using forward selection of clinically relevant variables and those with significance P < .10 in univariate analysis. Statistical analyses were performed using SAS 9.3 (SAS Institute, Cary, NC).

Results

A total of 150 patients met criteria for inclusion in the study. Of these, 76 (50.6%) had at least one episode of hypoglycemia, with 42 (28.0%) having clinically important hypoglycemia and 16 (10.7%) experiencing severe hypoglycemia during the admission. Mean glucose levels were lower in the group with hypoglycemia (159.6 vs 172.1 mg/dl, P < .001). The hypoglycemia group was younger (58.3 vs 62.6 years, P = .054), had a higher baseline HbA1c (7.8 vs 6.9, P = .017) and a lower BMI (29.3 vs 32.0, P = .045). The hospital length of stay and number of comorbidities were similar between the two groups (Table 1).

Table 1.

Patient Characteristics: Overall and With and Without Hypoglycemia.

Variable	Overall (n = 150)	Hypoglycemia (n = 76)	No hypoglycemia (n = 74)	P value
Age in years	60.4 ± 14.0	58.3 ± 15.5	62.6 ± 12.0	.054
Weight in kg	89.2 ± 26.3	85.7 ± 25.0	92.8 ± 27.2	.099
Height in cm	169.9 ± 10.9	169.6 ± 11.4	170.2 ± 10.5	.762
BMI, kg/m2	30.6 ± 8.1	29.3 ± 7.0	32.0 ± 8.9	.045
Albumin in g/dl	3.27 ± 0.59	3.23 ± 0.57	3.31 ± 0.61	.408
Male sex	79 (52.7%)	42 (53.2%)	37 (46.8%)	.519
Race				.303
Caucasian	96 (64.9%)	53 (55.2%)	43 (44.8%)
African American	42 (28.4%)	17 (40.5%)	25 (59.5%)
Asian	2 (1.4%)	1	1
Hispanic	1 (0.7%)	1	0
Other	7 (4.7%)	2	5
Comorbidities	4.7 ± 1.5	4.9 ± 1.5	4.5 ± 1.5	.182
Diabetes				.003
Type 1	28 (18.7%)	23 (82.1%)	5 (17.9%)
Type 2	115 (76.7%)	49 (42.6%)	66 (57.4%)
Unknown	5 (3.3%)	3	2
CFRD	2 (1.3%)	1	1
Steroids				.373
None	132 (88%)	66	66
<5 mg	2 (1.3%)	2	0
5-10 mg	16 (10.7%)	8	8
Hospital length of stay in days	20.0 (range 2-242)	23.9 ± 39.9	16.0 ± 22.7	.135

Continuous variables presented as mean (SD). Categorical variables presented as n (%).

Table 2 summarizes glucometrics in the overall cohort and also by whether hypoglycemia occurred. The group with hypoglycemia received a higher total daily dose (TDD) of insulin by weight of 0.34 units/kg compared to 0.23 units/kg in the group without hypoglycemia (P = .003). Mean basal insulin by weight was significantly higher in the hypoglycemia group (0.198 units/kg vs 0.136 units/kg, P = .01). Basal to total insulin ratios were similar between the groups.

Table 2.

Comparison of Glucometrics and Insulin Doses in Those With and Without Hypoglycemia.

Variable	Overall (n = 150)	Hypoglycemia (n = 76)	No hypoglycemia (n = 74)	P value
HbA1C % (n = 91)	7.33 ± 1.82	7.76 ± 2.2	6.87 ± 1.16	.017
Lowest BG, mg/dl	73.0 ± 30.2	50.9 ± 12.2	95.8 ± 25.9	<.0001
Highest BG, mg/dl	313.8 ± 97.8	328.8 ± 106.8	298.4 ± 85.5	.057
Average pre-HD glucose, mg/dl	176.9 ± 56.6	173.0 ± 62.2	180.8 ± 50.2	.399
Standard deviation of pre-HD glucose	47.8 ± 29.1	51.5 ± 31.1	43.9 ± 26.6	.111
Average post-HD glucose, mg/dl	167.3 ± 47.8	161.0 ± 50.8	173.7 ± 44.0	.102
Standard deviation of post-HD glucose	48.8 ± 29.5	55.0 ± 47.9	42.4 ± 36.2	.009
Clinically significant hypoglycemia (BG < 54 mg/dl)	42 (28%)	42 (55.3%)	0	<.001
Severe hypoglycemia (BG < 40 mg/dl)	16 (10.7%)	16 (21.1%)	0	<.001
Total insulin, units	25.3 ± 22.9	28.8 ± 24.3	21.6 ± 20.9	.054
Total insulin by weight, units/kg	0.288 ± 0.234	0.344 ± 0.247	0.231 ± 0.206	.003
Basal insulin, units	15.1 ± 15.7	17.2 ± 15.2	12.9 ± 16.1	.097
Basal insulin by weight, units/kg	0.168 ± 0.161	0.198 ± 0.156	0.136 ± 0.162	.019
Basal: total insulin	0.568 ± 0.342	0.604 ± 0.302	0.531 ± 0.377	.196

Patients with type 1 diabetes formed 19% of the total cohort with 82% experiencingat-least one episode of hypoglycemia compared to 42.6% in patients with type 2 diabetes (P = .003). A higher proportion of patients with type 1 diabetes had clinically significant hypoglycemia compared to patients with type 2 diabetes (Table 3). Although mean glucose was similar between type 1 and type 2 diabetes (163.5 mg/dl vs 166 mg/dl, P= 0.467), glucose variability was higher in type 1 diabetes (Table 3).

Table 3.

Comparison of Glucometrics in Type 1 and Type 2 Diabetes Patients.

Variable	Type 1 DM (n = 28)	Type 2 DM (n = 115)	P value
BMI, kg/m2	28.35 ± 6.57	31.54 ± 8.37	.063
HbA1c % (n = 91)	7.9 ± 2.5	7.1 ± 1.4	.143
Lowest BG, mg/dl	55.5 ± 18.0	78.1 ± 31.5	<.001
Highest BG, mg/dl	365.8 ± 126.1	301.1 ± 87.3	.015
Average pre-HD glucose, mg/dl	188.8 ± 70.8	173.3 ± 52.2	.287
SD of pre-HD glucose	66.4 ± 31.3	43.2 ± 25.5	<.0001
Average post-HD glucose, mg/dl	174.7 ± 57.9	166.1 ± 45.5	.402
SD of post-HD glucose	67.7 ± 40.8	44.8 ± 24.7	.008
Hypoglycemia (BG < 70 mg/dl)	23 (82.1%)	49 (42.6%)	<.001
Clinically significant hypoglycemia (BG < 54 mg/dl)	14 (50%)	25 (21.7%)	.003
Severe hypoglycemia (BG < 40 mg/dl)	6 (21.4%)	10 (8.7%)	.055
Total insulin, units	25.1 ± 15.5	25.3 ± 24.6	.957
Total insulin by weight, units/kg	0.32 ± 0.19	0.28 ± 0.24	.372
Basal insulin, units	15.0 ± 11.4	15.1 ± 16.4	.972
Basal insulin by weight, units/kg	0.18 ± 0.13	0.17 ± 0.17	.689

The majority (61%) of hypoglycemic episodes occurred on pre-HD days and fewer (14.6%) occurred within the 8 hours immediately after HD. A third of patients had early morning hypoglycemia, accounting for 37% of all episodes of hypoglycemia. Among all early morning hypoglycemic episodes, a majority (63%) occurred on the morning of the dialysis day, prior to the HD session. The total insulin dose used on pre-HD and post-HD days was similar (24.9 vs 23.8 units). In patients who received short acting meal or correction insulin within 6 hours prior to the HD session, there was a significant drop in mean glucose by 46 mg/dl post-HD (95% CI 30.3-62.5, P < .001). The glucose drop was 6.9 mg/dl in patients who did not receive any short-acting insulin pre-HD.

When exploring thresholds for hypoglycemic risk, we observed a linear relationship between TDD of insulin and hypoglycemia. At all cutoff points between 0.10 and 0.30 units/kg, higher TDD was associated with a statistically significant 2 to 3 times odds of hypoglycemia. A cutoff of 0.20 units/kg, had 64.5% sensitivity and 54.1% specificity for hypoglycemia (Table 4).

Table 4.

Test Characteristics of Various Cutoff Points of Total Daily Insulin Dose in Predicting Hypoglycemia.

Cut off for TDD (units/kg)	Sensitivity	Specificity	Odds ratio (95% CI)	P value
>0.05	96.1	10.8	2.95 (0.75-11.6)	.147
>0.10	89.5	25.7	2.94 (1.20-7.22)	.019
>0.15	76.3	44.6	2.59 (1.29-5.22)	.008
>0.20	64.5	54.1	2.14 (1.11-4.11)	.023
>0.25	51.3	68.9	2.34 (1.20-4.55)	.013
>0.30	44.7	81.1	3.47 (1.66-7.25)	<.001
>0.35	34.2	82.4	2.44 (1.14-5.24)	.022

Odds ratio is for risk of hypoglycemia in patient with TDD above cutoff versus below cutoff.

In multivariate analyses, patients with type 1 diabetes had 7 times higher odds of developing hypoglycemia compared to type 2 diabetes. After adjustments for significant factors, insulin dose (total or basal) was no longer a significant predictor of hypoglycemia (Table 5).

Table 5.

Logistic Regression Model to Predict Hypoglycemia.

Variable	OR	95% CI	P value
HbA1C (per 1.0 change)	1.22	0.86-1.71	.263
Age (per 1.0 change)	0.98	0.94-1.01	.181
Total insulin by weight (per 0.1 change)	1.06	0.76-1.49	.726
Basal insulin by weight (per 0.1 change)	1.27	0.77-2.10	.354
Type 1 diabetes	6.83	1.68-27.7	.007

Discussion

This is the first study evaluating glucometrics and clinical factors associated with hypoglycemia in hospitalized patients with diabetes and ESRD. Over 50% of patients developed at least one episode of hypoglycemia with 10% experiencing glucose values <40 mg/dl and 28% <54 mg/dl. Multiple episodes of hypoglycemia were noted in 35% of patients. Type 1 diabetes had much higher rates of hypoglycemia than those with type 2 diabetes and ESRD (82% vs 43%). These rates are much higher than the published inpatient rates in patients without ESRD (19-23%)^18,19 and indicate the vulnerability of this population. They are similar to comparative groups of outpatient ESRD study populations though. In a 3-month outpatient observational study of 102 ESRD patients on oral medication and insulin, the incidence of hypoglycemia was reported to be 52.9%.²⁰ This high incidence is attributed to the inability of the failed kidney to clear exogenous insulin and oral hypoglycemia agents leading to longer half -lives of diabetes medications. In addition, the response to hypoglycemia is attenuated by impaired renal gluconeogenesis.⁹

Our results support using conservative insulin dosing for glycemic management of patients with ESRD on HD. Due to the recognized risk of hypoglycemia, expert recommendations suggest using lower TDD of insulin in CKD patients (GFR < 60 ml/min) ranging from 0.2-0.3 units/kg body weight.¹⁴ Even within this range we observed a significant incidence of hypoglycemia. The ACP provides guidance to decrease insulin dose by 50% of the initial dose when GFR falls below 10 ml/min.¹⁵ There is sparse data to support these insulin dose recommendations for patients on HD. A prospective randomized controlled study of hospitalized diabetes patients with GFR <45 ml/min described a significant reduction in hypoglycemia rates (15.8% vs 30%, p 0.08) when the TDD of insulin was reduced to 0.25 units/kg, compared to 0.5 units/kg.¹¹ We similarly found a significantly higher mean TDD of 0.34 units/kg/day in the hypoglycemia group compared to 0.23 units/kg in the group without hypoglycemia. Although risk decreased with lower TDD of insulin, we did not identify a “safe” cutoff below which hypoglycemia risk was minimal. Therefore, clinicians should use a cautious and individualized approach.

In outpatient studies lower average glucose and increased risk of hypoglycemia have been noted in the post-HD period.^14,21 In a study of 10 outpatients the authors recommended a 25% reduction in basal insulin requirement on post-HD days.²² This reduced requirement was attributed to improved insulin sensitivity post-HD. However, in our hospital study, hypoglycemia rates were higher pre-HD (especially in the am prior to the HD session) and only 15% of hypoglycemic episodes occurred in the immediate post-HD period. This difference in timing of hypoglycemia may be due to the inpatient nature of our study where insulin was the predominant diabetes medication. In excessive doses, insulin stacking prior to HD could lead to hypoglycemia. Clearing of insulin during dialysis could reduce this risk post-HD. Therefore the inability of kidney to clear insulin in hospitalized ESRD patients may lead to similar doses having differential effects pre- and post-HD.

Patients experiencing early morning hypoglycemia had a higher TDD of insulin in addition to a higher basal dose (ratio 0.62 vs 0.54), although this was not statistically significant. The lack of significance may be due to the small size of this group (51 patients). These results suggest that basal heavy insulin regimens may increase the risk of hypoglycemia in ESRD and a lower basal to bolus insulin ratio (≤ 0.5) may be more appropriate in this population. Collectively these data do not support the need to lower insulin doses post-HD, but rather suggest reduction of the TDD of insulin in hospitalized ESRD patients. We also observed that short acting insulin given within 6 hours prior to HD resulted in a significant drop in glucose post-HD, we therefore suggest using conservative doses at this time.

This study has several limitations including its retrospective nature and a small sample size in a single institution. There is considerable provider variability in managing diabetes and we were unable to control for diabetes treatment and adjustments or evaluate factors related to that. We also could not evaluate the impact of dietary timing and physical activity in relation to hypoglycemia. In our institution patients are not permitted to eat during an HD session so we were unable to evaluate the effect of prandial insulin given during HD on hypoglycemia risk. Glucose data in this study was by point of care or laboratory measurements. We did not have access to continuous glucose monitoring which has been shown to be helpful in studies of this nature. Despite these limitations, we were able to evaluate the risk of hypoglycemia in one of the largest published cohorts of hospitalized patients with diabetes and ESRD on HD. Our data corroborate findings from outpatient studies of similar populations and highlight the vulnerability of this population to glucose fluctuations.

Conclusion

Rates of hypoglycemia are very high in insulin receiving hospitalized patients with diabetes on HD. Type 1 diabetes patients had the highest risk. Conservative insulin dosing with a focus on reducing TDD rather than changing doses in relation to HD is recommended. Prospective inpatient studies using continuous glucose sensors can further refine these recommendations to reduce hypoglycemia in this vulnerable population.