Insulin resistance has a key role in type 2 diabetes (T2D). However, there may be differences in transcriptomic associations of these phenotypes within different studies. Here we re-analysed data from GSE27951 and GSE20950 microarrays from adipose tissue to understand the differences in differential expression, find overlaps in differentially expressed genes and understand classification performance. We retrieved pre-processed microarray data from GSE27951 and GSE20950 using Affymetrix Human Genome U133 Plus 2.0 arrays from the Gene Expression Omnibus . We performed a differential expression between T2D and NGT (n = 10 vs 11 samples) and insulin resistant vs sensitive subcutaneous adipose tissue (IR vs IS)(n = 9 vs 10 samples). 3 samples were removed from GSE27951 as they had conflicting annotations. We performed differential expression using a probe-wise welch test. We performed false discovery rate correction using the Benjamini Hochberg method. We performed a threshold on the absolute difference in expression (log2(1.5)). Logistic regression was performed with feature selection and scaling to explore classification performance. We performed additional analysis to account for sex, perform permutation tests to understand significance of classification and look at alternative annotation. No probes met the FDR threshold of < 0.05 in T2D vs NGT. 13,124 probes met the FDR threshold of < 0.05 in IR vs IS. Once adding the effect size threshold no probes met the criteria in T2D and 3,501 probes were selected. Once adjusting for sex 3,356 probes were selected in IR. Cross validated classification using feature selection in training gave AUC values of 0.818 and 1.000 in T2D vs NGT and IR vs IS respectively. However, once performing permutation tests on classification using 100 label shuffles we found empirical p-values of 0.0495 and 0.0099. Changing the 3 conflicting annotations in GSE27951 did not give any probes meeting the FDR in T2D. The two cohorts produced substantially different differential-expression results under the implemented analytical workflow. The large IR-associated signal persisted after adjustment for sex, but technical confounding and differences between cohorts remain unresolved. Although internal cross-validation produced high AUC values, particularly for IR, these findings do not establish external predictive validity or identify clinically validated biomarkers. Further independent validation is required.