Mass spectrometry provides possibilities to simultaneously measure multiple endocrine-related analytes to more accurately diagnose hormonal disorders than possible using any single analyte alone. The same multi-analyte panel may also be applied to different disorders. Thus, measurements of metanephrines and methoxytyramine can be used for diagnosis of phaeochromocytoma or paraganglioma (PPGL) and neuroblastoma, whereas a single mass spectrometry-based steroidomics panel can be employed for primary aldosteronism (PA), adrenocortical carcinoma, Cushing syndrome and other disorders of steroidogenesis. Moreover, patterns in test results can be used to assess post-test probabilities of disease and subtype disorders according to tumour location, disease-causing mutations or even malignant risk. The problem remains how to best interpret such multidimensional data. With today’s advances in computational power, this can be achieved using machine learning (ML) classification-based algorithms. Sufficiently large numbers of patients are crucial for training and internal validation, which should be followed by external validation. For training, correct classifications are essential, though problematic for some disorders. We have specifically developed ML models for PPGL and PA. For PPGL we have employed datasets of over 700 patients with and over 3200 without tumours to train and validate ML models for determination of post-test disease probabilities and assessments of metastatic risk. Predictions of ML-models out-perform those of specialists in the field. For PA we have developed ML models using a retrospective cohort of 462 tested patients, including 139 with unilateral disease with 58 due to KCNJ5 mutations. Those ML models are now undergoing external validation in a prospective study that has to date enrolled 542 eligible patients. Although use of our models for screening was compromised by incorrect classifications due to immunoassay inaccuracy, one model nevertheless performs similarly to aldosterone:renin ratios. Significantly improved performance was achieved by incorporation of plasma potassium and renin with that steroidomics-based model. Among the patients in the trial, 186 have received follow-up, including 38 with unilateral disease confirmed by post-surgical biochemical cure. Genotyping has identified KCNJ5 mutations in 13 patients, all correctly predicted by ML to have PA with KCNJ5 mutations. Ten had their adrenals removed without AVS evidence of lateralisation. Such patients with an identified adrenal mass may thus proceed directly to surgery without need for confirmation studies or AVS. Although applications of multidimensional diagnostics with artificial intelligence show promise for diagnostic stratification of endocrine disorders, it remains important to establish that outcomes for patients are improved compared to traditional procedures.