Machine learning-based algorithm as an innovative approach for the differentiation between diabetes insipidus and...

Area of Science:

• Endocrinology
• Medical Diagnostics
• Artificial Intelligence

Background:

• Differentiating central diabetes insipidus (cDI) from primary polydipsia (PP) is clinically challenging.
• The hypertonic saline infusion test aids diagnosis but is labor-intensive and requires close patient monitoring.
• Machine learning (ML) offers a potential solution for more efficient differential diagnosis.

Purpose of the Study:

• To develop and validate an ML-based algorithm for accurate differentiation between cDI and PP.
• To assess the diagnostic accuracy of ML models using various clinical, biochemical, and radiological data.
• To determine if ML can reduce the need for complex diagnostic procedures.

Main Methods:

• Analysis of data from 59 cDI and 81 PP patients from a prospective multicenter study.
• Identification of key differentiating covariates using standard ML techniques.
• Development and validation of ML algorithms on unseen test datasets.

Main Results:

• A basic ML algorithm using clinical and lab data (urine osmolality, plasma sodium/glucose, surgery history, pituitary deficiencies) achieved an AUC of 0.87.
• Incorporating MRI characteristics improved accuracy (AUC: 0.93).
• Adding hypertonic saline test results further enhanced diagnostic performance (AUC: 0.98).

Conclusions:

• ML algorithms can accurately differentiate cDI from PP using readily available clinical and laboratory data.
• This ML approach may obviate the need for cumbersome diagnostic tests like the hypertonic saline infusion test.
• ML facilitates a more efficient and potentially less invasive diagnostic pathway for cDI.