The main finding of this study is that there is substantial variation in the conduction of hyperinsulinaemic–hypoglycaemic clamps across research groups, particularly in terms of antecedent day preparation, IIRs, number of hypoglycaemic steps, hypoglycaemic nadirs and duration of the clamps. The methodology descriptions frequently lacked important information in that less than half of the articles provided all the information needed to evaluate the experiment. Clamps were usually performed in the morning, in fasting condition, with or without some form of standardisation of meals ingested the evening before the clamp. Also, most studies included instructions for people with insulin-treated diabetes to adjust insulin use to avoid (nocturnal) hypoglycaemia prior to clamping. Only one of ten studies imposed other lifestyle restrictions before the clamp, such as refraining from alcohol or caffeine intake, smoking or engaging in strenuous exercise, which can affect glucose homeostasis and responses to hypoglycaemia [9,10,11,12,13,14], although the duration of these restrictions varied from 12–24 h [15, 16] to 72 h [17, 18]. More than half of the experiments were done in participants without diabetes and only around 7% of participants had type 2 diabetes.
There was an almost 50-fold difference between the highest and lowest IIRs used during the clamps, not including the doubling of the insulin dose that some studies applied to reach the lowest glucose target in stepped clamps. In addition, many studies with participants with type 2 diabetes increased the IIR at the lowest glucose level to ensure it was maintained in the face of insulin resistance and a brisk counterregulatory response. Apart from its effect on glucose requirements, changing the ambient insulin level may affect outcomes, such as the response of counterregulatory hormones. The direction of this change is not known, with some studies observing lower counterregulatory responses of high-dose vs low-dose insulin  and others finding the exact opposite, albeit all in healthy men [20, 21]. Another study, performed in participants with type 1 diabetes, did not find a difference in counterregulatory hormone responses between high- or low-dose IIRs . It should be noted that even the lower insulin doses are often unphysiologically elevated, as this is needed to achieve hypoglycaemia.
Whether the IIRs are calculated on the basis of body weight or BSA does not seem to be relevant. Indeed, the CVs of achieved plasma insulin levels across participants as an estimate of inter-individual insulin variability did not reveal meaningful differences between the two calculation methods. It should be noted, however, that only very few studies included obese individuals, which is relevant because obesity has a much greater effect on the calculated insulin dose when this is based on body weight rather than on BSA. Indeed, for the abovementioned person (75 kg, BSA 1.94 m2), the calculated insulin doses for an IIR of 60 mU m−2 min−1 or 1.5 mU kg−1 min−1 are about similar (6.98 vs 6.75 U/h), yet when this person weighs 125 kg (BSA 2.41 m2), these doses equal 8.68 and 11.25 U/h, respectively.
An important indicator of the quality of a glucose clamp is the CV of achieved plasma glucose levels for each glucose step. The CV reflects the stability of the glucose levels achieved during the clamp; the lower the CV, the more stable the clamp. Although there is no formal consensus on how low the CV of glucose levels should be during a hypoglycaemic clamp, a CV <5% is generally considered desirable . However, this was achieved in only about a third of the articles, and it is plausible that CVs are worse in articles that neither reported nor provided the option to calculate the CV.
Soluble short-acting human insulin was used in the majority of the clamps, although some studies also used porcine insulin . Since rapid-acting insulin analogues and human insulin possess the same pharmacodynamic and pharmacokinetic qualities when administered directly into the bloodstream, the choice of insulin is not considered to affect the outcome of the experiment itself. In the early days of clamp history, a priming insulin dose was often administered to quickly achieve target insulin levels. However, due to the very short t½ of insulin, such a priming dose is probably unnecessary for insulin doses below ~2 mU kg−1 min−1 (~80 mU m−2 min−1)  and increases the risk that glucose levels fall too quickly. For higher insulin doses, a priming dose has been calculated to shorten the time until reaching steady-state glucose disposal in normoglycaemic glucose clamps .
Data on GIRs were not systematically reported. Ideally, they should be reported separately for each glycaemic phase. The GIR during hypoglycaemia is a surrogate marker of the combined counterregulatory hormone response, reflecting the inverse of endogenous glucose appearance resulting from hormonal counterregulation.
Most, but not all, clamps using venous blood sampling applied some form of hand warming to achieve arterialisation of venous blood. Because insulin stimulates glucose uptake in skeletal muscle, peripheral venous samples underestimate, to a variable degree, the glucose concentrations in the blood supplying tissues, most importantly the brain. Proper hand warming opens arteriovenous shunts, resulting in arterialisation of venous blood. Liu et al. found an arteriovenous difference for high and low IIRs of 0.9 ± 0.1 and 0.4 ± 0.1 mmol/l, respectively , whereas the arterial–arterialised venous blood difference was about 0.1 mmol/l (95% CI −0.2, 0.4) . The heated hand box method, by which the local temperature is raised in a controllable way to 55–60°C, is widely used to arterialise venous blood . However, the method by which blood is arterialised is less important, as long as the temperature is sufficiently elevated. Indeed, raising the temperature to 40°C with warm blankets was found to be equally effective as the heated hand box . It should be acknowledged that although the arterialisation method is reasonably well validated for glucose, this may not be the case for other compounds (e.g. counterregulatory hormones) , indicating that it is not possible to arterialise venous blood completely.
The vast majority of articles reported measurement of glucose levels in plasma, while the remainder of the articles reported these to be measured in whole blood. This is important, because, depending on the haematocrit, glucose levels are approximately 11% lower in whole blood than in plasma . Indeed, most point-of-care glucose meters use standard algorithms to convert glucose measured in whole blood to plasma glucose. Also, the haematocrit may not be stable during clamps, which introduces bias. There is a high risk of misinterpretation when data in studies are compared without considering the source of glucose measurement from either whole blood or plasma. This is particularly relevant for the determination of hypoglycaemic thresholds, e.g. for release of counterregulatory hormones and deterioration of cognitive function, which inform decisions on the cut-offs used in the current classification for hypoglycaemia .
In 2017, the IHSG proposed glucose levels <3.0 mmol/l (<54 mg/dl), coined as clinically important hypoglycaemia, to be reported in clinical studies, so as to enable comparing of the effectiveness of interventions with hypoglycaemia as an endpoint . The majority of the clamp studies that we investigated included a glucose level around this value, but about one of every four single-step clamps used a glucose nadir that was substantially higher (up to 4.3 mmol/l). The 3.0 mmol/l threshold level is the result of consensus and analyses are currently being conducted to refine and solidify the level . For comparability reasons, it could be argued to always include such a refined threshold value in future hypoglycaemic glucose clamps, whether involving one or multiple steps.
There was also substantial variability with respect to the duration of the hypoglycaemic steps used in both the single-step and the stepped clamps. Whereas a duration of hypoglycaemia as short as 5 min (at 2.9 mmol/l) has been reported to initiate the process of habituation , a common protocol in clamp studies is to take approximately 20 min to reach that level and another 20 min to revert back from hypoglycaemia. The CGM definition of hypoglycaemia requires such an event to last for a minimum of 15 min, with prolonged hypoglycaemia defined as an episode of at least 120 min . On the other hand, long duration of hypoglycaemia can be seen as highly unphysiological, affecting both the counterregulatory response  and potentially other outcomes. It seems plausible that the longer the duration of the hypoglycaemic phase, the more discomfort this may cause, so that a maximum duration of 30–60 min seems reasonable.
A total of 11 studies in this systematic review were performed in children (age range, 6.4–18.0 years), ten of which included children with type 1 diabetes. The methodology of the study protocols in the paediatric population was very similar to those of adult populations with respect to IIRs, glucose targets and overall duration. However, since the number of studies is small, extrapolating our findings in the adult population to children should be done with caution, particularly since the younger age group (<12 years) is underrepresented.
This review has limitations. Due to the large number of articles, some of which dated back >40 years, we only extracted information from the article itself and chose not to contact the authors. Also, we focused on type 1 and type 2 diabetes. Studies involving people with insulinomas , pancreatic transplantation , gastric bypass  or other conditions unrelated to diabetes were therefore excluded to minimise potential further methodological variability. Information about the use of albumin or the participants’ blood to prevent insulin from sticking to the infusion sets was very sparse. Similarly, very few studies provided information about the addition of potassium to the glucose/insulin infusion to avoid hypokalaemia and the potential arrhythmia-provoking consequences . However, the lack of such information may suggest these adverse events to be extremely uncommon. More than 60% of the articles lacked other important information, which may reduce the validity for assessing data and comparing studies.
In conclusion, there is substantial variation in how hyperinsulinaemic–hypoglycaemic clamps have been conducted and reported in the past >40 years. This variation may potentially impact or raise questions about the validity of outcomes, and certainly makes it difficult, if not impossible, to compare results across studies. International consensus to standardise the design of both single-step and stepped clamps is therefore urgently needed.