Limitations of insulin resistance assessment in polycystic ovary syndrome

Background Though insulin resistance (IR) is common in polycystic ovary syndrome (PCOS), there is no agreement as to what surrogate method of assessment of IR is most reliable. Subjects and methods In 478 women with PCOS, we compared methods based on fasting insulin and either fasting glucose (HOMA-IR and QUICKI) or triglycerides (McAuley Index) with IR indices derived from glucose and insulin during OGTT (Belfiore, Matsuda and Stumvoll indices). Results There was a strong correlation between IR indices derived from fasting values HOMA-IR/QUICKI, r = −0.999, HOMA-IR/McAuley index, r = −0.849 and between all OGTT-derived IR indices (e.g. r = −0.876, for IRI/Matsuda, r = −0.808, for IRI/Stumvoll, and r = 0.947, for Matsuda/Stumvoll index, P < 0.001 for all), contrasting with a significant (P < 0.001), but highly variable correlation between IR indices derived from fasting vs OGTT-derived variables, ranging from r = −0.881 (HOMA-IR/Matsuda), through r = 0.58, or r = −0.58 (IRI/HOMA-IR, IRI/QUICKI, respectively) to r = 0.41 (QUICKI/Stumvoll), and r = 0.386 for QUICKI/Matsuda indices. Detailed comparison between HOMA-IR and IRI revealed that concordance between HOMA and IRI was poor for HOMA-IR/IRI values above 75th and 90th percentile. For instance, only 53% (70/132) women with HOMA-IR >75th percentile had IRI value also above 75th percentile. There was a significant, but weak correlation of all IR indices with testosterone concentrations. Conclusions Significant number of women with PCOS can be classified as being either insulin sensitive or insulin resistant depending on the method applied, as correlation between various IR indices is highly variable. Clinical application of surrogate indices for assessment of IR in PCOS must be therefore viewed with an extreme caution.


Introduction
The term 'polycystic ovarian syndrome' (PCOS) represents a heterogeneous and multifaceted entity characterised by hyperandrogenism and/or ovulatory dysfunction. It is also the most common endocrinopathy of women of reproductive age (1,2). According to the Rotterdam criteria (2003) (3), a diagnosis of PCOS can be established when at least two out of three criteria are present (oligo-/anovulation, clinical hyperandrogenism or biochemical hyperandrogenaemia and polycystic ovaries) on condition that other causes of oligo-/ anovulation or hyperandrogenism/hyperandrogenaemia (hyperprolactinaemia, Cushing's syndrome, congenital adrenal hyperplasia, premature ovarian failure, hypothalamic/pituitary disease, etc.) have been ruled out.

Subjects and methods
The study included 478 women aged 24.75 ± 8.05 years (mean ± s.d.), body mass index (BMI) 27.27 ± 7.18 kg/m 2 who underwent investigations for irregular periods, hirsutism or biochemical hyperandrogenism in the Department of Endocrinology and Metabolic Diseases, The Medical University of Lodz and The Polish Mother's Memorial Hospital Research Institute in Lodz, Poland (between 2013 and 2016). A diagnosis of PCOS was established according to the Rotterdam consensus criteria (3). All patients were subjected to an identical investigation protocol that included hormonal assessment (TSH, free T 4 and free T 3 , prolactin, total testosterone, androstenedione, DHEAS, 17-hydroxy-progesterone, cortisol after 1 mg overnight dexamethasone suppression test, fasting blood lipids and intravaginal pelvic ultrasound). All subjects also underwent glucose and insulin measurements during 75 g OGTT, where measurements were performed at 0, 60 and 120 min.
If clinically indicated, additional tests (such as IGF-I, growth hormone during OGTT, 17-hydroxy-progesterone measurements after 250 µg of intravenous Synacthen, 24-h prolactin secretion profiles) were performed. All investigations were performed in the early follicular phase of either a spontaneous cycle or after induction of the menstrual bleeding with a progestagen (usually dydrogesterone (Duphaston) 10 mg twice a day for ten days).
Insulin resistance index (IRI) was calculated from changes of glycaemia and insulinaemia during a 75 g oral glucose tolerance test (OGTT) according to the method described by Belfiore and coworkers (9). IRI was calculated through the formula: ISI (Gly) = 2/(1/(INSp × GLYp)) + 1, where INSp and GLYp are the measured insulin and glycaemic areas. In normal subjects, ISI(gly) are always around 1, with maximal variations between 0 and 2. This method is based on changes of glycaemia and insulinaemia during OGTT. According to the same authors, the assessment of free fatty acids (FFA) during OGTT is equally effective for the purpose of calculation of the IRI (9).
As in fact there are several formulae used to calculate Stumvoll index (11), we have chosen the most commonly used two formulae: where: I 0 , fasting insulin (pmol/L); I 120 , insulin concentration at 120 min of OGTT (pmol/L); G 120 , glucose concentrations at 120 min of OGTT (mmol/L). 7:3 As inclusion of parameters, such as age and BMI, in our opinion, could enrich analysed models, based almost exclusively on glucose and insulin, then we have decided to include into our analysis also a formula for the Stumvoll index that involves few measurements during OGTT, but incorporates demographic data, such as age and BMI into the model: As patients previously diagnosed with type 2 diabetes according to high fasting blood glucose criterion (glucose concentrations >7.0 mmol/L) do not require an OGTT to confirm a diagnosis of diabetes, then they were not included into the study.
Statistical analysis was performed by the means of MedClac software, version 16.4.3. Clinical and hormonal characteristics of subjects participating in the study are presented in the study are presented in Table 1. The study was approved by the Ethics Committee of the Polish Mother's Memorial Hospital-Research Institute.
Consent has been obtained from each patient or subject after full explanation of the purpose and nature of all procedures used.

Results
During OGTT, 19 patients (3.97%) were found to have impaired fasting glucose (i.e. glucose concentrations 5.56-7.0 mmol/L), 42 patients (8.78%) were found to have impaired glucose tolerance (i.e. glucose concentration 7.0-11.1 mmol/L at 120 min of OGTT), while 5 patients (1.04%) were found to have frank diabetes (glucose concentrations >11.1 mmol/L at 120 min of OGTT). Four of these patients (0.8%) were also found to have simultaneously impaired fasting glucose. Coexistence of both impaired fasting glucose and impaired glucose tolerance was observed in eight (1.67%) patients.
Percentile distribution of IR indices calculated according the HOMA, IRI, QUICKI, McAuley index, Matsuda index and Stumvoll 0,120 and Stumvoll demographics models is presented in Table 2.
In further analysis, we have assessed concordance and discordance between selected IR indices. Due to a strong correlation between HOMA-IR and QUICKI indices (r = −0.999), only comparison between HOMA-IR and IRI was used for further assessment. We have also chosen comparison between HOMA-IR and IRI indices, as these indices are routinely used for assessment of IR in our department. Further analysis ( Fig. 2A, B and Tables 5, 6) revealed that a significant number of patients would be differently classified in terms of percentile distribution, according to the method applied. Hence, at 75th percentile, out of 132 patients found to be above 75th percentile for IRI, only 70 (53%) would be concomitantly found to be above 75th percentile according to HOMA-IR. The same persisted for 90th percentile, where only 44% of patients found to be above 90th percentile for IRI, was simultaneously above 90th percentile for HOMA-IR ( Fig. 2B and Table 6). These mentioned discrepancies tended to persist even at extremes of IR spectrum (i.e. above 95th percentile - Fig. 2C and Table 7). Interestingly, the above discrepancies were even amplified, where IRI was compared to the data obtained from McAuley index, where 121/126 (96%), and 52/53 (98%) women with IRI above 75th, and 90th percentile had the value of McAuley index below 75th and 90th percentiles, respectively (Tables 8 and 9). The above mentioned discordance also persisted for values above 95th percentile (Table 10).

Discussion
This study, based on analysis of one of the largest group of women with PCOS, diagnosed in a single centre, according to an identical protocol, leads to three main conclusions.    There are also data that correlation between HOMA-IR and insulin concentrations during OGTT in women with PCOS is relatively modest (for instance r = 0.42, and r = 0.52, at 60 and 120 min of OGTT respectively) (16). Furthermore, it is known that IR indices derived from fasting glucose and insulin predominantly reflect hepatic rather than peripheral insulin sensitivity that is more reflected by indices that are based on glucose and insulin during OGTT (17,18). As correlation between IR indices based on fasting vs OGTT-derived data are highly variable even if identical clinical data (e.g. fasting glucose and insulin) are included into the mathematical model (see HOMA-IR and QUICKI); hence, in our opinion, it is virtually impossible to select 'the best' surrogate method for the assessment of insulin resistance in women with PCOS.
In our study, we observed a significant, but relatively weak correlation between all analysed IR indices and total testosterone. Indeed more IR subjects seem to have higher testosterone/dihydrotestosterone ratio, and significant, though not particularly strong correlation with HOMA-IR and QUICKI was reported before (19). To the best of our knowledge, this is, however, the first study, where six different IR indices were correlated with serum androgens in such a large group of women with PCOS. Interestingly, we did not observe a correlation of IR indices with DHEAS that was reported by Brennan and coworkers (20), though those authors used only HOMA-IR model for their assessment.
From clinical perspective, it is important that significant discrepancies between the methods based on fasting values and OGTT-derived values seem to persist even at the extremes of insulin sensitivity spectrum, i.e. when analysed according to percentiles of data distribution. Hence, a significant number of women classified as most  insulin resistant according to one method/s (e.g. based on OGTT-derived data), might be found to be less (or more) insulin resistant according to a different method (e.g. based on fasting data), regardless of a percentile used as a cut-off point. Thus, if a 75th percentile is used, then 47% of women found to be insulin resistant by IRI, would fall below 75th percentile for HOMA. Due to an excellent correlation between HOMA-IR, QUICKI indices, we can assume that the same situation would apply, if QUICKI index were substituted for HOMA-IR. The same applies to 90th percentile, as well as to 95th percentile of data distribution. Interestingly, the opposite situation, i.e. the number of women found to be more insulin resistant according to HOMA-IR than IRI, appears to be less frequent (15%, and 6%, respectively, for 75th and 90th percentile). Given only moderate correlation between McAuley index and OGTT-derived indices (Matsuda, IRI (Belfiore) and Stumvoll) assessment IR by the means of an McAuley index should not be extrapolated to IR assessment based on glucose and insulin measurements during OGTT, as discrepancies are even greater at upper extremes of IR percentile distributions (Tables 8, 9 and 10). The last, but not the least, we can state, even in the absence of a control group, that women with PCOS seem to be more insulin resistant than the general population. For instance, 75th percentile for HOMA-IR is 3.25 for our population of women with PCOS, while 75th percentile for the Polish population of Krakow (Poland) was reported as 2.1 (21), and 2.29 for the Czech population (22). Most quoted cut-off point for the 90th percentile for HOMA-IR is 3.8 (23), though precise calculations, also for the Spanish population suggested a cut-off for HOMA-IR of 3.46 (24), at 90th percentile of data distribution. For comparison, a 90th percentile for HOMA-IR for our PCOS patients equalled 4.6. The same applies for the Insulin Resistance (Belfiore) Index, where cut-off points for the 75th and 90th percentile were 1.46 and 1.67, respectively, while the quoted cut-off point for this index (no percentile specified) is 1.27 (25). Mean values of HOMA-IR and QUICKI in our study, are, however, similar to data of Christodoulopoulou and coworkers (26), based on a group of 309 Greek women with PCOS. It must be remembered, however, that our data, though based on a large group of women with PCOS, have been obtained from an almost entirely Caucasian population, while percentile distribution for IR indices may be different, if derived from other (e.g. Asian) populations (27).
Finally, the results of our study should be interpreted in view of potential utility of IR assessment in women with PCOS. The issue of insulin resistance in PCOS, though seemingly obvious, is indeed highly problematic, when supposed to be transformed from a theoretical concept into a clinical application. In a seminal paper by Dunaif and coworkers (28), IR in PCOS was assessed by the means of euglycaemic glucose-clamp technique in a relatively small group of women, i.e. in nineteen obese and ten non-obese women with PCOS vs eleven obese and eight non-obese controls. The authors concluded that IR was apparent not in terms of surpassing of any predefined cut-off point, based on selected surrogate IR indices, but    In summary, analysis of our data, based on a large number of women with PCOS, demonstrated that fasting triglycerides can be safely used instead of fasting glucose for assessment of insulin resistance, when compared to methods utilising fasting glucose and insulin. On the other hand, application of methods based on glucose and insulin measurements during OGTT vs methods based entirely on fasting data yields discrepant results in terms of severity of insulin resistance in a significant number of women with PCOS, with a wide variation of correlation coefficients between various methods. Hence, women classified as insulin resistant by one method, might not be equally insulin resistant if analysed by another method. Therefore, in our opinion, currently it is not possible to select either a universal 'cut-off' point in order to define insulin resistance, nor to define 'the best' method of assessment of insulin resistance, for the purpose of clinical practice.

Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

Funding
The study was supported by the statutory funds from the Medical University of Lodz (503/1-107-03/503-11-001) and Polish Mother's Memorial Hospital -Research Institute, Lodz, Poland.