Diabetes During Diarrhea-Associated Hemolytic Uremic Syndrome : A systematic review and meta-analysis

The primary research questions were as follows. 1) What is the incidence of diabetes during acute D+HUS? 2) Is diabetes during acute D+HUS associated with more severe D+HUS, as defined by the presence of coma or seizures, requirement for dialysis, or death? and 3) What is the long-term prognosis of those who develop diabetes during acute D+HUS? Specifically, what proportion of children recover, have permanent diabetes, and relapse after initial recovery?

All language case series, cohort studies, and randomized trials that described patients with D+HUS (idiopathic, diarrhea, or E. coli O157:H7 associated) and examined patients for hyperglycemia or diabetes were included. Two reviewers independently evaluated eligibility of all studies. A third reviewer resolved disagreements.

Citations were retrieved from Medline, Embase, and the Institute for Scientific Information from 1966 to 28 February 2005. The search strategy was pilot tested and modified with known relevant articles in an iterative process and included the terms hemolytic uremic syndrome, thrombotic thrombocytopenic purpura, diabetes mellitus, glucose intolerance, hyperglycemia, glucose, insulin, pancreas, pancreatic diseases, pancreatic necrosis, extrarenal, and nonrenal. Reference lists of included articles were hand searched, and articles citing included articles in the Institute for Scientific Information index were reviewed.

To summarize the characteristics of the included studies, two investigators developed a methodological assessment tool based on criteria of internal and external validity, as previously described for studies of prognosis (9) (Table 1). Using standardized forms, two investigators independently rated each article on these criteria, as well as abstracted data on study, patient, and outcome characteristics. Disagreements were resolved by consensus.

The mean age, proportion of patients with various characteristics, and long-term outcomes are described for all children with diabetes and D+HUS. CIs for single proportions were obtained using the Wilson score method (10).

Those studies that described incidence of diabetes were included in the meta-analysis. Data from these studies are presented in tabular form. χ² Tests were used to assess between-study heterogeneity. The I² statistic, which describes the percentage of total variation across studies due to heterogeneity rather than chance, was calculated (11). Due to significant heterogeneity among studies, an approach based on generalized estimating equations, which accounted for within- and between-study variability (random-effects modeling), was used to derive pooled estimates and their variances (12). Overall, each study contributed a weight of between 3.7 and 11.7%. Estimates were computed using R2.0.1 (R Foundation for Statistical Computing, Vienna, Austria).

For studies included in the meta-analysis, the odds of having diabetes with severe D+HUS (i.e., central nervous system symptoms, need for dialysis) was compared with the odds of having diabetes without severe D+HUS; the odds of dying with diabetes was compared with the odds of dying without diabetes. As between-study heterogeneity was not significant for the individual odds ratios (ORs), the Mantel-Haenszel method was used to calculate fixed-effects pooled ORs for each of these three factors (13). All analyses were performed with complete data, and P values <0.05 were considered statistically significant.

Three hundred eighty-four citations in six languages were screened from all sources, and 93 full-text articles were retrieved for detailed review. Twenty-one articles met inclusion criteria (5,14–33). Seven articles seemed to describe patients already included in this review (34–40). Any additional data from these studies were combined with included studies. The agreement beyond chance between two independent reviewers for article inclusion was good (κ = 0.74).

Interrater reliability in the assessment of methodological criteria was good (overall κ = 0.65 [range 0.39–1.0]) (Table 1). No study met all criteria. Cohort studies met 3–11 of 15 criteria, while case series met 3–9 of 11 criteria.

Patients were classified as having diabetes based on the primary author’s assessment or if it was evident that the patient was treated with insulin. In one study (23), it was unclear whether the patients were treated with insulin, as the article just stated diabetes was present. Three stated “insulin-dependent diabetes” was present (5,21,26), and one stated “diabetic ketoacidosis” was present (22). Of the remaining 16 studies, 15 referred to patients with hyperglycemia with or without ketoacidosis treated with insulin (14–20,24,27–33); the glucose cut points defining hyperglycemia were specified in only one study (glucose >360 mg/dl [20 mmol/l] on 3 consecutive days) (24). One study defined diabetes according to standard international criteria (25).

Point estimates of the incidence of diabetes with D+HUS were reported in 13 of 21 studies (14–26); these were pooled using meta-analysis (Table 2). These 13 studies described 1,189 children aged 0.2–16 years, from five different countries, over the period 1966–1998. There were no adult studies meeting inclusion criteria. One study described an outbreak of D+HUS from meat contaminated with Shiga toxin–producing E. coli (17), 10 of 13 studies described sporadic D+HUS (14–16,18,19,21–25), and 2 studies described both outbreak and sporadic D+HUS (20,26). Follow-up was restricted to the acute phase of D+HUS (<3 months) in 6 of 13 studies (14–16,19,24,25) and to 12–48 months in 7 of 13 studies (17,18,20–23,26).

The incidence of diabetes in the individual studies ranged from 0 to 15% (Fig. 1). The variability among studies was larger than would be expected by chance (significant heterogeneity, P = 0.007), with I² = 61%. Given the high between-study variability, the results pooled using the random-effects model should be interpreted with caution. The pooled estimate of diabetes was 3.2% (random-effects model, 95% CI 1.3–5.1%). We recalculated the pooled estimates after excluding studies that were outliers or qualitatively different. The pooled estimates were similar when studies that did not define diabetes (23) or had some patients with secondary HUS (21,22,24) were excluded (3.4% [95% CI 1.4–5.5] and 3.9% [1.5–6.2], respectively). Two small studies (14,17) had the highest point estimates (8.1 and 15%); the pooled estimate was 2.3% (1.1–3.4) when these two studies were excluded.

The presence of diabetes was associated with more severe disease. Patients with central nervous system symptoms were more likely to develop diabetes than those without (OR 12.0 [95% CI 4.2–34.2], P < 0.00001, n = 471 patients in seven studies), as were patients who required acute dialysis versus those who did not (5.4 [1.6–18.4], P = 0.008, n = 355 patients in five studies). Patients with diabetes were also more likely to die than those without diabetes (15.0 [5.8–38.9], P < 0.00001, n = 712 patients in eight studies).

The 21 included studies described 49 children who developed diabetes during acute D+HUS (Table 3). All children except one developed diabetes within 14 days of presentation to hospital (mean 8 days [range 1–60]). Insulin was used in 86% of patients (n = 44), and ketoacidosis developed in 64% (n = 11). In at least one patient, hyperglycemia was noted before the initiation of peritoneal dialysis (18). Pancreatic autoantibodies, including islet cell, insulin, GAD, and insulinoma-associated protein 2 antibodies, were measured in five patients and found to be negative in all; methods used to measure these antibodies were not described (14,29–31,33). These five patients, as well as two additional subjects, were reported to have some measure of impaired insulin secretion (low endogenous insulin levels [14,25,28,31], low basal C-peptide levels [29], impaired response to Sustacal challenge [30], or abnormal first-phase insulin response to intravenous glucose [33]).

Long-term outcome was reported for 44 of 49 children. Of these 44, 23% died acutely (95% CI 13–37). In the 34 survivors, the mean follow-up was 32 months (range 0–180). Thirty-eight percent (95% CI 24–55) of survivors were left with persistent diabetes requiring insulin; 11 of 34 had persistent diabetes from the outset, while 2 of 34 redeveloped diabetes 3 and 60 months after initial apparent recovery. The remaining 21 children (62% [95% CI 45–76]) were reported to have complete recovery from diabetes, having required insulin for 19–120 days. However, follow-up was <12 months or not reported for 48% of these children.

Based on the available evidence, the incidence of diabetes (hyperglycemia requiring insulin) during acute D+HUS in children <16 years of age ranged from 0 to 15%, with pooled incidence estimated at 3.2% (95% CI 1.3–5.1). This is in contrast to the reported general population incidence of childhood type 1 diabetes of 0.013–0.040% (41–43). The development of diabetes was associated with severe disease, marked by central nervous system symptoms, the need for acute dialysis, and mortality. However, whether diabetes develops in the setting of milder disease is unknown, as all patients were not tested.

These findings would suggest frequent monitoring of blood glucose during acute D+HUS, especially for patients on peritoneal dialysis with glucose solutions. Early aggressive treatment of hyperglycemia will not only prevent ketoacidosis but may improve acute outcomes, as has been shown in other critically ill patients (44,45).

The pathophysiology of hyperglycemia during D+HUS is not known. Some have proposed that it is related to the initiation of peritoneal dialysis with glucose solutions or to the use of medications such as glucocorticoids or diazoxide. This is unlikely, as the hyperglycemic effects of diazoxide are transient (46); some patients never received glucocorticoids (18,19,25,26,28,30,31), diazoxide (18,19,25,26,29–31,33), or peritoneal dialysis (20,30); and one patient developed diabetes before peritoneal dialysis was instituted (18). It is possible that the children may have had “stress hyperglycemia” rather than true diabetes (47). However, stress hyperglycemia is usually transient (48), while here 38% of survivors required insulin at 12 months. Rather, the hyperglycemia may be due to insulin deficiency, as suggested by impaired insulin secretion in all eight patients who were tested (14,25,28–31,33).

The mechanism of insulin deficiency does not appear to be immune, as pancreatic autoantibodies were negative in all five patients studied (14,29–31,33). A viral etiology producing both diabetes and D+HUS is also unlikely, as viral studies were negative in one patient (31), viral particles were absent at autopsy in two patients (27,28), and the pathogenesis of D+HUS is not viral but related to Shiga toxin–producing bacteria (3). Given that absolute insulin deficiency is not silent, it is unlikely that diabetes predated D+HUS in all patients. The mechanism may instead be the direct result of HUS with thrombotic microangiopathy of the pancreatic microvasculature. Pancreatic necrosis was present in four of five children with diabetes who had autopsies (25,27,28); two of these children demonstrated extensive thrombosis of pancreatic arterides (25,28). That several patients had concomitant exocrine pancreatic dysfunction also suggests a vascular mechanism.

The reasons for predilection of the pancreas to toxin-mediated damage are unclear. Gb3 receptors, associated with toxin binding in the kidney (49), do exist in the pancreas, although not in high density (50). During acute D+HUS, these receptors may be upregulated, and islet cells may be more susceptible to hypoxic ischemic injury due to decreased ability to clear free radicals (51). There are likely other factors involved in the pathophysiology, as one patient with diabetes was described as having no detectable pancreatic injury at autopsy (14).

Of children who developed diabetes during D+HUS and survived, over one-third had permanent diabetes requiring insulin 6 months to 15 years after the acute phase, whereas approximately two-thirds were reported to recover. Given that relapse can occur up to 5 years after the acute illness (32) and that almost 50% of recovering children were not followed for >1 year, the rate of permanent diabetes may in fact be higher than reported.

The mechanism of relapse may be continued loss of islet cells, or alternatively, decreased islet reserve in the face of a predisposition to type 2 diabetes. If this is the case, then even those who were not identified to have hyperglycemia during the acute phase of D+HUS may be at long-term risk of diabetes due to subclinical islet cell injury. Further studies are needed to determine the long-term risk of diabetes in survivors of D+HUS. Until then, consideration should be given to screening all survivors of D+HUS for diabetes with an oral glucose tolerance test at least 1 and 2 years’ postillness.

The quality of a review is dependent on the quality of the original studies. Although all included studies used a cohort representative of patients with D+HUS at their institutions, the majority were retrospective, few reported a specific definition for D+HUS or diabetes, and long-term follow-up was variable. We only included articles mentioning diabetes, predisposing to potential selection of positive studies and overestimation of the true incidence. To ascertain the impact of such a publication bias, we determined that at least 90 studies, with 50 patients per study, would have to report no cases of diabetes in order to change the pooled incidence from 3.2 to <0.5%. Alternatively, this review may have underestimated the true incidence of diabetes, given that diabetes was not systematically assessed in any of the 13 studies.

Finally, it could be argued that a meta-analysis should not have been performed due to the substantial heterogeneity among studies, potentially resulting in an erroneous estimate of the true incidence of diabetes with D+HUS. The random-effects pooled estimate represents an approximation of the mean incidence, accounting for the variable precisions of the individual estimates of incidence and their pattern of variation. The precision of the pooled estimate takes into account the uncertainty in the estimates from the included studies, as well as the additional between-study variation observed. In keeping with this point, the pooled estimates calculated after excluding outliers fell within the 95% CI of the pooled estimate calculated using all included studies. Thus, the overall pooled estimate with 95% CIs allows us to succinctly summarize existing evidence. Ideally, we would have liked to perform metaregression to explore reasons for heterogeneity but were limited by inadequate statistical power. Potential reasons for the variability include differences in surveillance for diabetes (given that none of the included studies’ main purpose was to evaluate incidence of diabetes) and differing severity of D+HUS.

Despite its limitations, this review summarizes the best available evidence on the important but infrequently recognized complication of diabetes associated with D+HUS and highlights the need for further research.

Diabetes is an important complication of D+HUS. Early identification and aggressive treatment of hyperglycemia during D+HUS may improve outcomes acutely. Future studies that prospectively evaluate the incidence of this complication both during the acute phase of D+HUS and after long-term follow-up are needed. Meanwhile, children who present with D+HUS should have aggressive surveillance and treatment of hyperglycemia. Consideration should also be given to long-term screening of all survivors of D+HUS for diabetes.

This work was supported by the Canadian Diabetes Association and the Ontario Ministry of Health.

This study was performed in conjunction with the Walkerton Health Study. Other study investigators include Douglas Matsell, MD, Louise Moist, MD, MSc, and Marina Salvadori, MD.