Abstract

Methods

Results

A flowchart summarizing the literature search and screening results is shown in Figure 1. A total of 1360 abstracts were screened for inclusion in the study. Of these abstracts, 424 were excluded because they were not in the pediatric population or did not use HOMA to assess insulin resistance. The remaining 936 full-text articles were obtained and assessed for eligibility. Of these articles, 298 met the criteria for inclusion in the final analysis. Thirty-four studies were excluded at the full-text screening stage because they included patients older than 18 years, 5 were excluded because they were review articles or letters to the editor, and 9 were excluded because they used HOMA to classify patients as having metabolic syndrome. The remaining 590 articles were excluded because they measured HOMA in children but did not use a cutoff to define insulin resistance. Examples included comparing HOMA in different populations, tracking changes in response to an intervention, and correlating HOMA to other factors, such as body mass index.^24–26

Open in a separate window

Figure 1

Flowchart of study selection in a review of pediatric literature in which homeostasis model assessment (HOMA) cutoff values were used to diagnose insulin resistance.

Among the 298 studies included in this review, 51 different HOMA cutoff values ranging from 0.77 to 6.3 were used to classify patients as having insulin resistance (Figure 2). The most frequently used values were 3.16 and 2.5. Forty-three studies (14.4%) used a percentile cutoff specific to the study population. Of the 255 studies (85.6%) that used a predetermined fixed cutoff to define insulin resistance, 72 (28.2%) provided a reference that supported the use of that cutoff in the population. Forty-eight studies (18.8%) provided no reference for their cutoff values, and 62 (24.3%) cited a study that was irrelevant (did not discuss HOMA). Twenty-three articles (9%) cited a study that used the same cutoff value but did not validate it. In addition, 50 studies (19.6%) cited a reference for the HOMA cutoff that clearly did not support the use of that cutoff in their study population (Figure 3). For example, several studies cited the 1985 study by Matthews et al⁶ to support the 2.5 cutoff for defining insulin resistance. This study⁶ had a small population size, receiver operating characteristic (ROC) curves were not generated, and the authors did not propose 2.5 as a cutoff for diagnosing insulin resistance. That number seems to reflect the study’s finding that the median HOMA score for an overnight basal sample was 2.5 in the 6 adult diabetic participants compared with 1.3 in the 6 nondiabetic participants.⁶

Open in a separate window

Figure 2

Frequency and range of pediatric homeostasis model assessment (HOMA) cutoffs (N=298). Fifty-one different cutoff values ranging from 0.77 to 6.3 were used. The most frequently used cutoffs were 3.16 and 2.5.

Open in a separate window

Figure 3

Assessment of articles that used a fixed cutoff. The majority of studies used a predetermined cutoff value to diagnose insulin resistance (n=255). Citations supporting the use of that cutoff were subsequently evaluated.

The most commonly cited reference for the 3.16 cutoff was Keskin et al.¹³ This study compared HOMA to OGTT in 57 pubertal obese children and adolescents. The ROC analysis completed in that study identified 3.16 as the most appropriate cutoff in this population. However, numerous studies cited this article as support for using 3.16 in prepubertal populations despite the fact that insulin resistance is known to increase naturally during puberty.^10,11 The most commonly cited references for the 2.5 cutoff were Valerio et al²⁷ in 2006, Madeira et al²⁸ in 2008, and Matthews et al.⁶ As described above, the study by Matthews et al⁶ did not validate the use of 2.5 as a cutoff value in either adult or pediatric populations. The study by Valerio et al²⁷ reported the prevalence of insulin resistance in a population of obese children and adolescents in southern Italy. In that study, 2.5 was used as a cutoff for defining insulin resistance in children, and 4.0 was used as a cutoff for adolescents, citing a 2004 study by D’Annunzio et al.²⁹ Although the authors were contacted, we have been unable to obtain this article. However, an abstract³⁰ presentation and a later article by the same group³¹ reveal the likelihood that these numbers were based on percentiles of HOMA according to Tanner stage in a population of about 100 healthy children. The study by Madeira et al²⁸ used data from overweight prepubertal children to identify HOMA cutoff values for predicting metabolic syndrome. Although metabolic syndrome and insulin resistance are related, they are not equivalent. Using this study to validate the 2.5 cutoff for diagnosing insulin resistance is not appropriate. Other studies that attempted to establish HOMA cutoffs for identifying metabolic syndrome or assessing cardiovascular risk were also cited by some groups as evidence to support their use in diagnosing insulin resistance.^32,33

Discussion

Osteopathic medicine emphasizes the importance of disease prevention, which requires appropriate screening to identify patients at the highest risk of disease development and promote early intervention. Assessing obese and overweight pediatric patients for insulin resistance is an approach that may be valuable for the prevention of diabetes and other cardiometabolic diseases. However, standard methods for assessing insulin resistance in children have not been established. In clinical and epidemiologic studies, HOMA is a measurement frequently used to assess insulin sensitivity. Our current review of the pediatric research literature identified a great degree of inconsistency in how HOMA is being used to define insulin resistance and identified extremely limited evidence to support even the most commonly used HOMA cutoff values. Although the difference between 2.5 and 3.16 may not seem significant, the choice of a cutoff value can have a large impact when determining the prevalence of insulin resistance in a population or identifying factors that contribute to its development. For example, in our previous study³⁴ of overweight and obese Appalachian children, 82% of the patients would have been classified as insulin resistant using 2.5 as a cutoff value, whereas the use of 3.16 would have resulted in a prevalence of 69%. These differences greatly complicate comparisons between studies and impede the development of practice guidelines.

Although inconsistent use of HOMA cutoffs was not surprising, the finding that so few of the selected cutoffs were supported by evidence was unexpected. More than 40% of the articles that used a predetermined cutoff either had no reference for the cutoff or cited an irrelevant study. Many authors simply cited a study that had used the same cutoff but did not seem to review the previous work to ensure that the cutoff was valid for their study population. When tracing back the citation history, many studies that selected 2.5 as a cutoff were ultimately basing its use on the original study by Matthews et al,⁶ which described HOMA as a measure of insulin resistance. This study clearly did not attempt to identify a cutoff point, and it provides insufficient information to support widespread use of a specific value for diagnosing insulin resistance in children or adults.

Our findings cast doubt on the quality of study design and adequacy of peer review. Failure to systematically review the literature before designing a study contributes to research waste.³⁵ This failure seems to be exhibited in the case of HOMA cutoff values. In the training of novice investigators, the importance of working through a citation history to validate elements of study design instead of simply following another group’s method should be emphasized. This type of mentorship requires the continued support of research opportunities for osteopathic medical students as we prepare them to advance osteopathic-focused research in the future.

The findings of the current review are limited by the fact that other minimally invasive surrogate measures of insulin resistance, such as the Quantitative Insulin Sensitivity Check Index, were not included. It is possible that the evidence supporting cutoff values for those indices is more substantial. Our review was also limited to articles available in the English language and indexed in PubMed. This approach was used to ensure inclusion only of articles published in journals that have been reviewed by the Literature Selection Technical Review Committee to confirm that they follow best practices, including peer review.³⁶ Additional evidence supporting the use of specific HOMA cutoffs may be available in excluded journals or unpublished studies. However, as previously noted by Rössner et al,¹⁵ the absence of standard methods for assessing insulin resistance continues to complicate interpretation of and comparisons between studies, thus contributing to research waste. The need to choose a standard surrogate measure and then conduct large population studies to establish sex- and age-specific ranges remains. Given the large number of published studies that assessed HOMA in pediatric patients, it may be possible to generate population ranges for HOMA or other fasting glucose and insulin-based surrogate measures through the sharing of raw data. This approach is more likely to provide clinically useful information than additional studies that compare HOMA or other surrogates with the OGTT or euglycemic-hyperinsulinemic clamp.

Conclusion

Studies using HOMA to diagnose insulin resistance in children used a wide range of cutoff values, most of which are not supported by evidence. These findings imply that inadequate review of the literature before study design is not uncommon and emphasize the importance of training new investigators to validate design components. The results also highlight the need to standardize methods for assessing insulin resistance in children. Given the large number of studies in which HOMA was measured, it may be possible to generate sex- and age-specific population ranges for HOMA or other surrogate measures based on insulin and glucose levels through data sharing.

Acknowledgments

Footnotes

References