Explainable Insulin Decision-making System to Assist Physicians in Diabetes Management

Completed

• Conditions: Diabetes
• Interventions: Other: without AI assistance; Other: with AI dosage assistance; Other: with explainable AI assistance; Other: with faulty explainable AI assistance

• Registration Number: NCT06434584
• Lead Sponsor: Shanghai Zhongshan Hospital

Brief Summary

The investigators plan to conduct a multi-case, multi-reader observational study with the primary objective of exploring the effects of an interpretable insulin-assisted decision-making system on physicians' (1) decision accuracy and (2) decision confidence.

Detailed Description

Not available

Study Timeline

• Study Start: 2022-12-10
• Primary Completion: 2023-03-20
• Study Completion: 2023-06-10
• Status Verified: 2024-05
• Study First Submitted: 2024-05-09
• Study First Submitted QC: 2024-05-23
• Last Update Submitted: 2024-05-23
• Study First Posted: 2024-05-30
• Last Update Posted: 2024-05-30

Recruitment & Eligibility

• Status: COMPLETED
• Sex: All
• Target Recruitment: 48

Inclusion Criteria

• Patients with T2DM who were admitted to Zhongshan Hospital of Fudan University from 2022.12 to 2023.3.
• Insulin regimen is one of the following: a. Basal regimen: once a day subcutaneous injection of long-acting or ultra-long-acting insulin; b. Premixed regimen: two/three times a day subcutaneous injection of premixed insulin; c. Basal mealtime regimen: three times a day before meals subcutaneous injection of short-acting or rapid-acting insulin, plus once a day injection of long-acting or ultra-long-acting insulin.

Exclusion Criteria

• Cases will be excluded if there is insufficient information for a valid assessment (missing data on insulin or blood glucose >40%).

Doctor

Inclusion Criteria:

licensed medical practitioner.

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Arm && Interventions

Group	Intervention	Description
senior doctor group	with AI dosage assistance	-
senior doctor group	with faulty explainable AI assistance	-
junior doctor group	without AI assistance	-
junior doctor group	with faulty explainable AI assistance	-
junior doctor group	with AI dosage assistance	-
senior doctor group	without AI assistance	-
senior doctor group	with explainable AI assistance	-
junior doctor group	with explainable AI assistance	-

Primary Outcome Measures

Name	Time	Method
Doctors' decision-making accuracy	up to 2 months	The investigators used MAE and clinical agreement to quantitatively assess the accuracy of clinician-recommended insulin doses.; 1. Mean Absolute Error (MAE) represents the error between the clinician-recommended value and the expert-recommended value (gold standard).; 2. Clinical agreement: Clinical agreement is calculated as the proportion of clinically consistent insulin doses (clinician-given adjustments in the same direction as the expert's protocol and within 20% of the dose difference) to the total insulin dose.
Doctors' decision-making confidence	up to 2 months	Clinicians' confidence in decision-making was assessed using a 10-point Likert scale from 1 (not at all confident) to 10 (completely confident).

Trial Locations

Locations (1)

Xiaoying Li; 🇨🇳 Shanghai, Shanghai, China