This study showed that women with PCOS complied with a low GI diet. There was a significantly lower mean GI of food items and GL of food items and meals in women randomised to the low GI arms of the trial compared to the healthy eating arm. The results suggested that compliance decreased as the study progressed although the mean GI and GL of food items and GL of meals were lower at all stages in the low GI group compared to the healthy eating group. The average GI of food items was 8.8% lower in the low GI group. The proportion of low, medium and high GI foods also differed significantly between the two arms, and the intake of high GI foods was lower in the low GI group. As far as we know, this was the first study to have assessed compliance to a low GI diet by analysing the GI value of data prospectively collected in food diaries and there were no published studies to compare the findings with.
The study was limited by various factors. A key limitation was the small sample size but although the study had a small number of women entered, the majority of diaries were well completed and the data available were large, so overall statistically significant differences were observable. However trends and sub group analysis were not statistically significant due to the small sample size. Although selection bias was limited due to randomisation, the small sample size may have increased the potential effect from volunteer bias and non-participation bias. Of those volunteering or referred to the trial only 19 met all eligibility criteria, 11 entered the trial and nine completed the trial. It is more likely that women who dropped out of the trial would not have complied with the dietary intervention, increasing the chance of the results showing compliance. However, a strength of the study was that it was linked to a rigorously conducted CRUK pilot which had consistent entry criteria, thorough randomisation, and good dietetic support for participants. The diaries were set out in a way encouraging a high level of detail, potentially allowing all food and drink consumed each day with quantities to be recorded. Printed recording booklets for food intake prompted patients for the desired information and structured data in an organised way facilitating data analysis. This assessment method, when completed properly, was a robust way of gathering data and has been shown to have a beneficial reactivity effect  increasing compliance.
Another limitation was the lack of universal agreement on the GI values of foods, whether drinks such as tea and coffee should be included and the complexities around how to account for issues such as ripeness of fruit and specific combinations of foods which potentially affect each other. No account of food interactions was included in the analysis. The inclusion of standard portion sizes could have introduced inaccuracy but this will not have affected the results related to GI of food items recorded.
Although this study was single blinded, information bias could have occurred. The dietician knew which study arm patients were allocated to and more importantly the patients knew what intervention they were having in terms of low GI or healthy eating diet. In addition, the self monitoring by patients meant that control of data collection was the patient's full responsibility so the accuracy of the data relied on the patient's compliance to keeping the diary. The potential for bias in self completed diaries where the individuals knew what intervention they should be following was high. Less desirable eating episodes may have been excluded from the diaries, biasing the monitored behaviour in the desired direction. There was also the possibility of recall bias where information may have been entered retrospectively from memory leading to inaccurate recordings. The Hawthorne effect could have introduced bias during the whole study but particularly during the four weeks out of the six month trial that patients were required to fill in a food diary and the diaries may not have been representative of the other 20 weeks the patients were expected to comply with the dietetic advice. The direction of these biases would be to increase the likelihood of finding compliance to a low GI diet but the five percent weight loss in both arms of the trial suggests that the calorie deficit and/or exercise component was complied with. However it is not possible to totally rule out the Hawthorne effect in behaviourally based studies or to truly blind the participants to dietetic interventions.
The internationally accepted range for low GI intake is 0-55 and both groups in the study had an average GI of food items and GL of foods that was low. The average GI for items was 33.67 for the low GI arm and 36.91 for the healthy eating group. The average GL of items for the low GI diet was 8.15 and 9.81 for those on the healthy eating diet and it is suggested that the GL of items is low when under or equal to the value of 10 . These effects may be the result of the general advice and information given by the dieticians as many of the healthy eating diet foods suggested, such as salads, fruit and vegetables were similar to those suggested for the low GI diet and usually have a low GI. The main high GI foods within the diaries were potato and certain breads and breakfast cereals of which the participants of the low GI diet were advised to avoid in the personal record booklet suggesting all participants followed dietetic advice.
After enquires to find an appropriate programme to assess the food diaries it became apparent that an affordable commercial database was not available. The NutriGenie software initially looked a possible solution for qualitative analysis of whether the diet was predominantly low GI. However there was poor agreement of food classification when comparing NutriGenie and the SPSS database which used nationally published and accepted values (Kappa = 0.316). NutriGenie, despite claiming thousands of entries, contained significantly fewer foods from the diaries. It is not surprising that the commercial programme contained less relevant foods than the SPSS database as the latter was purpose made, but the difference in allocation of a food group to low, medium and high was surprising. The company marketing the NutriGenie software was reticent in giving information about sources of information for their database, but this is an illustration of the complexity of analysing the GI of diets accurately and also gives an understanding of the view of some health professionals that it is too complex to be a basis for dietary intervention.
This study, although small, is of interest due to the potential benefits of a low GI diet for the treatment of PCOS associated diabetes and obesity and the increasingly strong suggestion that the endocrine and metabolic abnormalities present in PCOS produce an association with endometrial cancer [1, 2, 5, 9].
A low GI diet where reduction of insulin levels lowers testosterone levels, improves hirsutism and acne, improves menstrual function, dislipidaemia and potentially decreases the risk of endometrial cancer [1, 11, 13–15] has led to support for its use in both obese and lean patients with PCOS . Realisation of any long term benefits such as cancer prevention would require compliance to the low GI diet. This study suggests that compliance is possible over a six month period, although longer term compliance would still need to be assessed. The benefits of lifestyle intervention in diabetics  show that a slightly restricted but healthy lifestyle can reduce long term health problems linked with insulin resistance and this study suggests that it might be possible for women with PCOS.