[Paper Review] Adherence to Personal Health Devices: A Case Study in Diabetes Management

Sudip Vhaduri, Temiloluwa Prioleau|arXiv (Cornell University)|May 30, 2020

Mobile Health and mHealth Applications33 references4 citations

TL;DR

This study uses data mining on 44 diabetes patients' continuous glucose monitor (CGM) data over 60–270 days to investigate adherence factors. It finds that longer data gaps—especially in abnormal glucose ranges—predict poorer outcomes, with 33% of gaps occurring during suboptimal glycemic control and the longest gaps in very low/high glucose states, particularly among poorly-controlled patients.

ABSTRACT

Personal health devices can enable continuous monitoring of health parameters. However, the benefit of these devices is often directly related to the frequency of use. Therefore, adherence to personal health devices is critical. This paper takes a data mining approach to study continuous glucose monitor use in diabetes management. We evaluate two independent datasets from a total of 44 subjects for 60 - 270 days. Our results show that: 1) missed target goals (i.e. suboptimal outcomes) is a factor that is associated with wearing behavior of personal health devices, and 2) longer duration of non-adherence, identified through missing data or data gaps, is significantly associated with poorer outcomes. More specifically, we found that up to 33% of data gaps occurred when users were in abnormal blood glucose categories. The longest data gaps occurred in the most severe (i.e. very low / very high) glucose categories. Additionally, subjects with poorly-controlled diabetes had longer average data gap duration than subjects with well-controlled diabetes. This work contributes to the literature on the design of context-aware systems that can leverage data-driven approaches to understand factors that influence non-wearing behavior. The results can also support targeted interventions to improve health outcomes.

Motivation & Objective

• To investigate how performance toward glycemic targets influences adherence to continuous glucose monitors (CGMs) in diabetes management.
• To examine the relationship between data gaps (missing CGM data) and blood glucose control outcomes.
• To identify whether non-adherence patterns correlate with diabetes control status (well-controlled vs. poorly-controlled).
• To inform the design of context-aware, data-driven systems that improve adherence and health outcomes in chronic disease management.
• To support targeted interventions by identifying behavioral patterns linked to suboptimal glucose management.

Proposed method

• Collected and analyzed 60–270 days of CGM data from 44 subjects with diabetes across two independent datasets.
• Defined data gaps as periods of missing or incomplete glucose readings, assuming these indicate non-adherence.
• Classified glucose levels into categories: normal, target, and abnormal (including very low and very high), using ADA glycemic target criteria.
• Computed average data gap duration per glucose category and compared across diabetes control groups (well-controlled vs. poorly-controlled).
• Used statistical analysis to assess the significance of associations between data gap duration and glucose level severity.
• Applied a data-driven approach to link adherence behavior (measured via data gaps) to clinical outcomes (glycemic control).

Experimental results

Research questions

• RQ1 To what extent does performance toward target glycemic goals influence wearing behavior of continuous glucose monitors?
• RQ2 How does the duration of data gaps vary across different glucose level categories (normal, target, abnormal, very low/high)?
• RQ3 Is there a significant difference in average data gap duration between patients with well-controlled and poorly-controlled diabetes?
• RQ4 Are data gaps more prevalent during periods of suboptimal glucose management, particularly in extreme glucose ranges?
• RQ5 Can data-driven analysis of missing data events inform context-aware interventions to improve adherence and health outcomes?

Key findings

• Approximately 33% of data gaps occurred when users were in abnormal blood glucose categories, indicating non-adherence during suboptimal management.
• Longest data gaps were observed in the most severe glucose categories—very low and very high—suggesting users discontinue CGM use during critical health states.
• Subjects with poorly-controlled diabetes had significantly longer average data gap durations than those with well-controlled diabetes.
• Data gaps were more frequent and prolonged when users were farthest from target glucose levels, highlighting a feedback loop of non-adherence during poor control.
• The study confirms that adherence to CGMs is strongly influenced by glycemic control performance, with non-adherence most pronounced during extreme glucose events.
• These findings support the development of context-aware systems that detect and respond to non-adherence patterns using real-time data gaps and glucose trends.