Exploration of patterns predicting renal damage in patients with diabetes type II using a visual temporal analysis laboratory

OBJECTIVE To analyze the longitudinal data of multiple patients and to discover new temporal knowledge, we designed and developed the Visual Temporal Analysis Laboratory (ViTA-Lab). In this study, we demonstrate several of the capabilities of the ViTA-Lab framework through the exploration of renal-damage risk factors in patients with diabetes type II. MATERIALS AND METHODS The ViTA-Lab framework combines data-driven temporal data mining techniques, with interactive, query-driven, visual analytical capabilities, to support, in an integrated fashion, an iterative investigation of time-oriented clinical data and of patterns discovered in them. Patterns discovered through the data mining mode can be explored visually, and vice versa. Both analysis modes are supported by a rich underlying ontology of clinical concepts, their relations, and their temporal properties. The knowledge enables us to apply a temporal-abstraction pre-processing phase that abstracts in a context-sensitive manner raw time-stamped data into interval-based clinically meaningful interpretations, increasing the results' significance. We demonstrate our approach through the exploration of risk factors associated with future renal damage (micro-albuminuria and macro-albuminuria) and their relationship to the hemoglobin A1C (HbA1C ) and creatinine level concepts, in the longitudinal records of 22 000 patients with diabetes type II followed for up to 5 years. RESULTS The iterative ViTA-Lab analysis process was highly feasible. Higher ranges of either normal albuminuria or normal creatinine values and their combination were shown to be significantly associated with future micro-albuminuria and macro-albuminuria. The risk increased given high HbA1C levels for women in the lower range of normal albuminuria, and for men in the higher range of albuminuria. CONCLUSIONS The ViTA-Lab framework can potentially serve as a virtual laboratory for investigations of large masses of longitudinal clinical databases, for discovery of new knowledge through interactive exploration, clustering, classification, and prediction.