Self-medication errors with insulin pens and inhalers can affect disease control but artificial intelligence (AI) systems that detect inappropriate use could help to correct technique and improve outcomes.
Poor adherence to drug therapy represents a major healthcare problem across the world and is estimated to average only 50% in developed countries. Moreover, reduced adherence can be further worsened by the use of devices such as insulin pens or inhalers for respiratory diseases. In fact, one study of those with chronic obstructive airway disease found that over half of patients experienced problems, irrespective of the device used. Although clinicians can provide instruction on the appropriate use of a device, patients may forget the correct technique over time so that disease control deteriorates. With recent developments in computer technology, a team from the Computer Science and Artificial Intelligence Laboratory, Massachusetts, US, wondered if an AI system could be used to assist patients in their homes and provide a continuous assessment of their administration technique and feedback when an individual failed to use the device correctly. The AI system was embedded in a sensor within the patient’s home, without the need for cameras or wearable sensors and performed an analysis of radio wave reflections from the environment to track the specific movements associated with medicine self-administration (MSA). The system was refined to detect whether a patient has followed all the required steps for using a medicine device. The researchers trained the system to focus on actions related to the correct sequence of events required for the use of insulin pens and inhalers.
Findings
The AI system was able to correctly detect an insulin pen administration event with a sensitivity of 87.6% and a specificity of 96.1%. Similarly, inhaler administration events had a sensitivity of 91.1% and a specificity of 99.2%. However, the researchers were more interested in the ability of the AI system to detect other actions related to the administration process. They specifically focused on not priming the insulin pen prior to use and not holding it still for 10 seconds after injection. For inhalation devices, the team looked at whether an inhaler was shaken before use and if the patient held their breath after inhalation. In all cases, there was a high degree of sensitivity and specificity for these measures. Furthermore, all of the information recorded is stored in the system cloud and fed back to the patient, providing data on the time that a dose was administered and whether the technique was correct. The patient was also given reminders if they failed to take a dose of treatment.
The authors concluded that their study was an important first step towards enabling automatic MSA assessment at home. However, they also note that further work is required to determine coverage of the system throughout the home and to address the effect of potential confounders such as dexterity issues and health literacy.
Citation
Zhao M et al. Assessment of medication self-administration using artificial intelligence. Nat Med 2021 https://www.nature.com/articles/s41591-021-01273-1
