1. Are the objectives, scope, and methods for this review clearly described?
Yes For the most part the scope/methods are clearly articulated and relatively easy to follow. A couple minor points that may warrant clarification in the methods:
  1. Though the results clearly delineate how each study defined severe hypoglycemia, I did not see the review methods specify how you were defining “severe hypoglycemia” for the purposes of study selection – I got the sense from results that you were very inclusive and left the definitions up to each study, but this would be worth stating explicitly in the methods. I also inferred from results that study had to essentially report incidence of symptomatic hypoglycemia – again, worth stating in methods. Also, what if the study did not explicitly define “severe hypoglycemia” but rather just presented incidences of glucose < 40 or < 60 or < 70? I assume these studies would be excluded because there was no mention of symptoms/need for assistance?
  2. What is the rationale for excluding studies of duration < 6 mos? Severe hypoglycemia is not really a time-dependent phenomenon (though the consequences of it may be). In any case, this is probably a moot point given the supplemental search, but may be worth more clearly defining rationale here. Also, the KQ1extension” is not mentioned in the methods, but then is presented in flow diagram – this may be confusing for readers and may want to include “extension” rationale and methods in the Methods section.
Introduction – small point – the exec summ background paragraph states intensive control only associated with reduction in microalbuminuria while the introduction in body of paper more properly states the broader impact of intensive control (esp since these include UKPDS) on other microvascular outcomes.
Analytic framework – the one thing that seems to be missing from this is patient behaviors – certainly things like exercise, inconsistent meals, medication mishandling etc would contribute to risk. I doubt these things are identified in any of the included studies, but the lack of such evidence may still be important to know about.
We moved the definition of severe hypoglycemia to the Methods section. We chose to exclude studies with fewer than 500 subjects and less than 26 weeks’ duration for feasibility; as it is we abstracted 60 studies for KQ1. As suggested, we included the rationale and methods for KQ1-extension in the Methods Section. We revised the executive summary background and the analytic framework as recommended.
No Although this dichotomous question requires a yes/no answer, neither is really correct. The review fails to put the issue of hypoglycemia in proper context. There is considerable variation in the definitions applied in studies of hypoglycemia. This variation and controversy surrounding it is important background. In addition, although a very explicit definition of severe hypoglycemia was chosen, there is a serious limitation as far as answering the Key Question #1: What is the incidence of clinically significant hypoglycemia? Their definition of severe hypoglycemia chosen was: “an episode with typical symptoms (e.g., sweating, dizziness, tremor, visual disturbance) that resolves after treatment (oral carbohydrate, intramuscular glucagon, or intravenous glucose) administered by another person.” There is clinically significant hypoglycemia that does not meet this definition. In addition, it does not address the issue of hypoglycemia unawareness which can result in unrecognized and untreated hypoglycemia with levels of glucose <40 mg/dl. (Compare reported rates to those reported on CGMS)We agree that there is clinically significant hypoglycemia that does not meet our definition and that asymptomatic low blood sugar (e.g., hypoglycemia unawareness) is not accounted for in this definition; however this is the definition that we chose based on its common use in the literature and that was approved by our TEP. We have acknowledged this point in our discussion.
2. Is there any indication of bias in our synthesis of the evidence?
Yes While there is no bias in selection of studies, from my perspective the report does not sufficiently emphasize the rates of serious hypoglycemia and possible morbidity/mortality for patients who are treated in the control arms of clinical studies or from observational data. For example, rates of potentially serious hypoglycemia in insulin treated patients was 59% in a study from a large HMO (Sarkar, 2010, Question 1). The association of serious hypoglycemia and morbidity/mortality from the standard arms of ACCORD/VADT/ADVANCE. Although observation data is not of as high quality, there are strong signals of high rates and potential harms in the selected VA populations which are not incompatible with patient self reported data. These issues are commented upon in section 4.Although it was included in KQ3, we realized that Sarkar et al. 2010 should have been included in KQ1 ext and added it. Thank you.
Yes I understand that large trials are needed to detect outcomes (i.e. severe hypoglycemia) that occur relatively infrequently. However, there were many trials with 400–499 patients with T2DM that reported the incidence of severe hypoglycemia. Some of these trials were part of the drug development program for the agent. What was the reasoning behind selecting the 500 patient cut-off? I am concerned that omitting these trials could introduce bias?See previous page, first response.
Yes Although this dichotomous question requires a yes/no answer, neither is really correct. My concern the way the results are presented and the use of the word “low” as in the following: “Overall incidence of severe hypoglycemia was low in the vast majority of the 60 reviewed studies, particularly those of metformin (0–1.5%), glucagon-like peptide-1 GLP-1 analogs (< 1%), dipeptidyl-peptidase-4 (DPP-4) inhibitors (<1%), insulin detemir (<1%), insulin aspartame (<1%), glinides (0%) and thiazolidinediones (TZDs) (<1%). Annual rates of severe hypoglycemia were greater than 1% for sulfonylureas and the following insulin preparations: neutral protamine Hagedorn (NPH), glargine, lispro and glulisine.”
“Low” is in the eye of the beholder. When up to 18% of patients on insulin report an episode of hypoglycemia requiring assistance in the previous year, that doesn’t sound low.
I do, however, appreciate consideration of additional studies “to gain a broader population-based perspective on incidence of symptomatic hypoglycemia.”
We agree that use of the term “low” to describe the frequency of severe hypoglycemia is a value judgment and we have either removed or modified that term in the final report.
3. Are there any published or unpublished studies that we may have overlooked?
Feil DG, Rajan M, Soroka O, Tseng CL, Miller DR, Pogach LM. Risk of hypoglycemia in older veterans with dementia and cognitive impairment: implications for practice and policy. J Am Geriatr Soc. 2011 Dec; 59(12):2263–72. Epub 2011 Dec 8. (rates of coded hypoglycemia in Veterans with cognitive impairment or dementia Seaquist ER, Miller ME, Bonds DE, Feinglos M, Goff DC Jr, Peterson K, Senior P; for the ACCORD Investigators. The Impact of Frequent and Unrecognized Hypoglycemia on Mortality in the ACCORD Study. Diabetes Care. Rhoads GG, Orsini LS, Crown W, Wang S, Getahun D, Zhang Q. Contribution of hypoglycemia to medical care expenditures and short-term disability in employees with diabetes. J Occup Environ Med. 2005 May; 47(5):447–52. Diabetes Care. 2012 Feb; 35(2):409–414. Epub 2011 Dec 16.
We thank the reviewers for bringing these articles to our attention. Of these, 3 were published after November 2011 which is when our last literature search was performed (Bonds, Feil, Seaquist); 2 had been excluded due to the fact that severe hypoglycemia was not defined (Raz, Swinnen); one we had already included (Rhoads), one was a duplicate publication of a study already included (Miser); one was a study of a newer agent approved by the FDA after our study was initiated (Owens); two meet our criteria, were not previously reviewed and have been added to our final report in KQ1 (Nauck, Russell Jones).
I randomly selected a few of the drugs (lispro, detemir, linagliptin, and liraglutide) and searched PubMed to see if there were other relevant articles. I came across the following articles that were >500 patients, ≥ 6 months, and presented data on severe hypoglycemia. It is not clear to me why these studies were excluded.
Raz I, et al. Effects of prandial versus fasting glycemia on cardiovascular outcomes in type 2 diabetes: the HEART2D trial. Diabetes Care. 2009 Mar;32(3):381–6.
Miser WF, et al, Randomized, open-label, parallel-group evaluations of basal-bolus therapy versus insulin lispro premixed therapy in patients with type 2 diabetes mellitus failing to achieve control with starter insulin treatment and continuing oral antihyperglycemic drugs: a noninferiority intensification substudy of the DURABLE trial. Clin Ther. 2010 May;32(5):896–908.
Swinnen SG, et al. A 24-week, randomized, treat-to-target trial comparing initiation of insulin glargine once-daily with insulin detemir twice-daily in patients with type 2 diabetes inadequately controlled on oral glucose-lowering drugs. Diabetes Care. 2010 Jun;33(6):1176–8.
Owens DR, et al. Efficacy and safety of linagliptin in persons with type 2 diabetes inadequately controlled by a combination of metformin and sulphonylurea: a 24-week randomized study. Diabet Med. 2011 Nov;28(11):1352–61.
Russell-Jones D, et al. Liraglutide Effect and Action in Diabetes 5 (LEAD-5) met+SUStudy Group. Liraglutide vs insulin glargine and placebo in combination with metformin and sulfonylurea therapy in type 2 diabetes mellitus (LEAD-5 met+SU): a randomised controlled trial. Diabetologia. 2009 Oct;52(10):2046–55.
Nauck M, et al. Efficacy and safety comparison of liraglutide, glimepiride, and placebo, all in combination with metformin in type 2 diabetes. Diabetes Care 2009; 32: 84–90.
See comment above.
No It is not specified in methods whether or not long-term consequences of inpatient hypoglycemia are considered an included study or not, but there is a study looking at long-term outcomes in patients who had had inpatient hypoglycemia: Svensson AM, McGuire DK, Abrahamsson P, Dellborg M. Association between hyper- and hypoglycaemia and 2 year all-cause mortality risk in diabetic patients with acute coronary events. Eur Heart J. 2005;26:1255–61.This article was not included because it focused on inpatients.
  1. More recent reports from ACCORD should be included, notably the ACCORD-EYE study and the ACCORD-MIND study, which showed reduction of retinopathy and reduction of brain shrinkage with intensive control of type 2 diabetes.
  2. Include the 3 year results of the 4T study: Holman RR et al. NEJM 2009;361:1736–47
  3. In addition to the report by Zoungas on associations of hypoglycemia with mortality risk, consider: Kosiborod M et al. JAMA 209;301:1556–64 and Boucai L et al. Am J Med 2011;124: 1028–35
We have reviewed all the articles mentioned, none of which met our criteria for inclusion (Kosiborod, ACCORD-EYE and ACCORD-MIND) or had already been included (4T Holman). Some of these, however, have been included in the discussion.
4. Additional suggestions or comments
From my perspective, the literature supports the following logic sequence that is relevant to VHA patient safety issues which I do not believe come thru in recommendations of the report.
  1. Based upon randomized trials of medications, most of which are industry funded and of shorter duration, serious hypoglycemia is uncommon, even in insulin treated patients.
  2. The recent ACCORD, VADT, ADVANCE studies were consistent in that while serious hypoglycemia was more common in the intensive arm, the health impact was greater in the standard arm for cardiovascular morbidity, and mortality (Zoungas NEJM 2010, Bonds DE BMJ 2010, Davis SJ (abstract, 2009), as well as with increased medical assistance (Miller et al BMJ 2010). The adjusted strength of association in the standard group in Accord was 2.87 (1.73 to 4.76); ADVANCE death from a cardiovascular cause (hazard ratio, 2.68; 95% CI, 1.72 to 4.19), VADT is not published, but the OR for recurrent severe hypoglycemia and mortality was 3.7. Although the recent article by Bonds et al (2012) found that prior episodes of serious hypoglycemia attenuated the association between hypoglycemia and mortality, it did not do so in the control arm. While it is not likely that this issue will even be conclusively resolved, the reviewer concludes that hypoglycemia is a strong risk factor for cardiovascular death in patients who are not “intensively treated”
  3. The risk factors for serious hypoglycemia are varied and differ across the studies, but include other medical conditions, minority status, neuropathy, cognitive impairment, limited health literacy. Although causality of hypoglycemia upon adverse outcomes cannot be proven, the results from the 3 major trials would clearly indicate that Veterans at high risk for serious hypoglycemia can be identified.
  4. The studies underestimate the risk of severe hypoglycemia in general practice, particularly for insulin treatment. A surveillance studies in an HMO (Sarkar 2010) noted that 59% of patients on insulin reported a significant hypoglycemia within a year. The Budnitz 2010 study, which will be included after review, will underscore that insulin and sulfonylurea remain high risk medications in the elderly. As noted, the Veteran literature is limited, but renal disease and cognitive impairment are two highly prevalent conditions associated with coded hypoglycemia; other factors, such as decreased health literacy, are also likely to be common in the Veteran population.
The ESP did identify Moen et al. as an article documenting an association with biochemical hypoglycemia and death in Veterans with CKD. Additionally, other studies (see section 2) indicate high rates of coded hypoglycemia in Veterans with coded hypoglycemia on insulin, and in an insured population on insulin. The rates of up to 17% cited in the conclusion of Key Question 1 may thus underestimate the rates in high risk populations on insulin therapy in both insured and Veteran populations.
Most of these excellent points have been included in our revised discussion.
In several places, insulin aspart is written as insulin aspartame. Insulin aspartame is incorrect and should be corrected so that it reads insulin aspart.
For the DPP-4 inhibitors, studies using vildagliptin were included (p. 95, 130–131); however, this product is not FDA approved.
In the Insulin glargine (primary therapy) studies, 4/5 allowed the patient’s prior oral diabetes medications to be continued (only Rosenstock 2001 did not allow concomitant oral agents). Therefore, these 4 trials were not truly primary therapy studies.
On p.126 Table 3b, Buse 2011 is listed under A. Regular Insulin and Lispro Studies; Fast-short Acting. The lispro used in this study was the 75/25 mix, which is an intermediate and fasting acting mixture so it should be listed under C. Biphasic Insulin: Intermediate and fast-acting mixture.
As suggested, we changed “aspartame” to “aspart”. Although vildagliptin is not FDA approved, it does appear in some of our tables because it was included in some of the studies that also used FDA approved agents.
The Buse study is now listed under “C” on Table 3B, as suggested.
Nicely done, thorough report.
My main suggestion has to do with the statement “Overall incidence of severe hypoglycemia was low in the vast majority of the 60 reviewed studies…”. Though this is true, it is somewhat misleading because the subsequent summary statements do not delve into the issue of glucose targets enough. If the achieved HbA1c in 58/60 studies were 7.5% or 8% in the intervention group, the low incidence of hypoglycemia in the vast majority of studies doesn’t really mean too much and it may suggest to readers that the bulk of evidence suggests that severe hypoglycemia is infrequent. I think the intensity of control really matters here and should be more clearly emphasized. It is hard to figure out from results and tables how the glucose target and/or glucose achieved relates to hypoglycemia incidence. Consider also saying more about the intensive vs less intensive evidence base in the summary statements/exec summary. Also, it might be useful to include the glucose targets for each of the studies in Table 3.
P18 – the NPH v glargine meta-analysis results are interesting. Many clinicians consider using glargine to help minimize hypoglycemia risk from NPH. I know this is not the focus of this paper, but the finding that the two drugs had equivalent risk of hypoglycemia has potential clinical importance and you could consider highlighting this more. Also, this is a pretty broad CI – I’m not sure I would say “risk is slightly higher” but not statistically significant – would probably just say no significant difference.
As suggested, we included an additional column in Table 1 (formerly Table 3) specifying the A1C targets and commented more extensively on the issue of intensive control in the executive summary, the summary statement, and the discussion.
We amended the statement regarding NPH vs glargine to indicate that the risk was not different, as recommended.
This is a well done review of hypoglycemia from the Evidence Based Synthesis Program ESP of the V.A. The goal of ESP Centers is to generate evidence synthesis on clinical practice topics and develop clinical policies informed by evidence guide the implementation of effective services to improve patient outcomes and set the direction for future research.
The current report examines in great detail the data available on hypoglycemia in adults with type 2 diabetes. The study is well done and provides a complete, well documented compilation of current information on severe hypoglycemia and will be a major resource for investigators in the area. It will also be of use in clinical care of patients in the V.A. The methods used in the study are appropriate and comprehensive. The study will be a very useful compilation of data on hypoglycemia for future clinical studies and will be of use in defining future directions. It has some limitations in its use by non-investigators in that the limitations of the various studies are not as well delineated in an easily accessible manner for the non-expert.
Many of these limitations are mentioned throughout the document, but it would be much more useful to the routine reader to have these limitations defined and a summary to help to better evaluate the data. As a simple example, many of the studies examining hypoglycemia in randomized control trials (RCTs) are obtained from pharmaceutical studies whose purpose is to establish non-inferiority of their agent against other agents in a very highly selected population. This is mentioned in the document, but again that could be lost for someone who does not read every word in the document. Another example is the use of superficially similar excellent studies, but directed at different populations and for different reasons to come to a single conclusion. One of the best examples of this are the ACCORD and ADVANCE trials, two of the best studies done on treatment of patients with type 2 diabetes but directed at different populations for different purposes. The ADVANCE study consisted of relatively mild diabetes with very few of the patients on insulin and low A1cs and ACCORD with a much more difficult population with almost half of the patients on insulin and much higherA1cs at the initiation of the study. The ACCORD trial had higher hypoglycemic numbers and consequences of treatment that may have been related to hypoglycemia which were quite detrimental.
Some of these issues of concern for the reader could be addressed in an additional summary of the limitations as mentioned above of individual studies. Another limitation of the current presentation is the difficulty in extracting clinical guidelines for care. While mentioned in the study, the clinical results in terms of outcomes of studies with high hypoglycemic rates may not justify the risk of very intensive control and perhaps standards of care could be qualified to include the risk of complications of treatment more clearly in the guideline.
A few specific comments: Some agents used for treatment of patients with type 2 diabetes, rarely if ever cause hypoglycemia when used as individual agents in patients without severe complications. The report clearly defines most of these including metformin, DPP-4 inhibitors, glinides, etc. Some of the insulins have not been extensively tested in routine use for example detemir data are mostly derived from pharmaceutical studies carefully designed to limit the risk of hypoglycemia. Other agents such NPH or glargine have much real world data and appear to be much riskier. For true risk of hypoglycemia with agents that do not typically cause hypoglycemia, it could be useful to include studies that use these agents in combination with the hypoglycemic agents such as insulin. This might give a better view of the risk in the usual use of these agents.
Minor Comments
A few typographical errors are present in the manuscript, the most glaring of which is on page 4 under Conclusions-an incomplete sentence is somewhat confusing.
Overall this is an extremely useful, carefully done, and valuable document for dissemination to professionals in practice and to researchers who will be planning future studies. I highly endorse this document and believe that it will be of great use in the V.A. and outside the V.A. for other practitioners and scientists.
Thank you.
We have summarized the limitations of the data in the executive summary and the discussion.
  1. Page 1 para 2: Microvascular complications other than albuminuria have indeed been shown: see the ACCORD-EYE study report in NEJM
  2. In Key Question #2 and elsewhere: glycated Hb is usually abbreviated as HbA1c, not HgbA1c.
  3. Page 3 para 1: Here and elsewhere insulin aspart is incorrectly referred to as ‘aspartame.’ Aspartame is an artificial sweetner; aspart is an insulin analogue. If the computer search was done with ‘aspartame’ it is no wonder no significant hypoglycemia was found. It cannot be concluded that aspart does not cause severe hypoglycemia or that it differs from other rapid acting insulin analogues in this way. An excellent report including data on hypoglycemic risk with aspart is: Holman RR et al. NEJM 2009;361:1736–47. Furthermore, the main prandial insulin used in the ACCORD trial was aspart, and in the intensive arm of this trial the incidence of events requiring medical assistance was greater than 3% yearly.
  4. Page 4, para 2: Here and elsewhere, ‘data’ is a plural noun.
  5. Page 9 bullet point 6: Why was gliclazide excluded from analyses? The ADVANCE trial is one of the best sources of information on long-term hypoglycemic risks, and it used gliclazlide. This drug is widely used throughout the world.
  6. Page 9 bullet point 3: A crucial point is glossed over here. Studies were included if they reported severe hypoglycemia, but there are wide variations between studies in both definitions of severe events and (just as important) ascertainment of such events. This is the main limitation of this analysis.
  7. Page 20 para 1: Ramadan is incorrectly spelled ‘Ramadam.’
  8. Page 21 last section: This summary statement reports annual incidence of severe events greater than 1% for NPH, glargine, lispro, glulisine, and sulfonylureas. Notably missing are aspart (a leading cause of severe events in ACCORD), premixed insulin (a leading cause of events in 4T and possibly the main cause of severe events in clinical practice), and regular insulin (certainly a leading cause of events when used in sliding scales in hospital, but not tested in big clinical trials and therefore missing from this analysis). Somewhere the probably causes of these omissions should be discussed.
  9. Page 41 next to last para, which reads: “It is also possible that the robust recent findings that intense glycemic control results in a more than two-fold increase in risk of severe hypoglycemia without any clear outcomes benefits, may lead to an appropriate relaxation in HgbA1c goal levels by both clinicians and guideline developers.” This statement should be amended in several ways. First, some guidelines are currently available which make the point that altering the A1c goals is appropriate for some patients, but not others. These actual guidelines should be cited for balance to this speculation. Also, the statement that there are no “clear outcomes” is incorrect. In ADVANCE and VADT, microalbuminuria was reduced. In ACCORD, microalbuminuria, retinopathy, and brain shrinkage were all reduced. In the long-term followup of UKPDS, all-cause mortality was reduced 27% in addition to microvascular events.

We have re-worded the executive summary to reflect the benefits of tight control on a variety of microvascular complications


All HgbA1C have been changed to HbA1C


The verbs accompanying the noun “data” are now in the plural form


As per our pre-determined methodology, gliclazide was not included since it is not an FDA approved medication


Our discussion points out that definitions and ascertainment of hypoglycemic events varied between studies and ascertainment may have been incomplete


We have corrected the spelling for Ramadan

5. Are there any clinical performance measures, programs, quality improvement measures, patient care services, or conferences that will be directly affected by this report? If so, please provide detail.
Insulin was identified as a high risk medication within VHA in the high alert medication group, with a final report issued in 2009. More recently, there has been renewed discussion in OSC, PBM, and some VISNs about the need to identify Veterans who at higher risk for hypoglycemia in order to decrease potential over treatment and to improve care coordination (e.g. telehealth, post hospital discharge) for those with identified events.
Pharmacy Benefits Management Services (PBM) along with the Medical Advisory Panel and VISN Pharmacist Executives are responsible for determining formulary status and guidance for use for pharmaceutical agents in the VA. The PBM would need to be made aware of any policies that would result from this report.
This summary could well affect the nature of diabetes performance measurement.
An important result of this report might be the design of prospective and structured collection of data to address the questions incompletely answered by this review of heterogenous data.We have included this point in our discussion.
6. Please provide any recommendations on how this report can be revised to more directly address or assist implementation needs.
As noted in comment 4, the reviewer recommends that the report give greater prominence to concerns that serious hypoglycemia is an identified risk factor for morbidity in and morality in “non-intensively treated subjects” from ACCORD, ADVANCE and VADT with mean achieved A1cs of 7.5%–8.4%; rates based upon survey and administrative data indicate incidence of potential serious hypoglycemia up to 59%; and that risk factors for hypoglycemia are not uncommon among the Veteran population.
See above responses to 1 and 2.
  1. This analysis and report are carefully done and generally confirm the findings of earlier efforts, including some important recently published data. However, the important limitations of the methods necessarily used should be included in the report.
  2. One such limitation is that the endpoint in question (hypoglycemia) is rarely the primary endpoint of clinical studies, and in many cases it is not a secondary endpoint either, just an occasionally reported safety observation. Application of rigorous meta-analytic methods cannot overcome this limitation of the data provided.
  3. Another limitation is that only some of the therapeutic agents commonly used have been included in the large, structured trials selected for this analysis. Hence, data are not available for drugs of interest. Regular insulin, for example, is a leading cause of hypoglycemia but its relative importance cannot be assessed using the present methods.
  4. Two other agents which pose significant risk of severe hypoglycemia also cannot be addressed by the present methods for similar reasons: the sulfonylurea glyburide, and all forms of premixed insulin. Hypoglycemia.
  5. Because of the limitations of the evidence available, few firm conclusions are possible. Rather, most of the observations are hypothesis-generating. Hence, a leading conclusion from this report should be that collection of better data, using the excellent VA data-handling system, would be very helpful.
We have included most of these points and limitations in our discussion.


Cover of Predictors and Consequences of Severe Hypoglycemia in Adults with Diabetes - A Systematic Review of the Evidence
Predictors and Consequences of Severe Hypoglycemia in Adults with Diabetes - A Systematic Review of the Evidence [Internet].
Bloomfield HE, Greer N, Newman D, et al.
Washington (DC): Department of Veterans Affairs (US); 2012 Apr.

PubMed Health. A service of the National Library of Medicine, National Institutes of Health.