Application and Development of Large Language Models in Smart Inhalers

Abstract

The emergence of generative artificial intelligence and Large Language Models (LLMs) has brought revolutionary applications in the medical field, especially in the field of smart inhalers, where LLMs show great potential. LLMs can optimize the functionality of smart inhalers, enhance patient education and feedback mechanisms, and support personalized medical decision-making through natural language processing and deep data analysis. However, the application of these technologies also presents numerous challenges. This paper systematically reviews the prospective applications of LLMs in smart inhalers, discusses the advantages of LLMs in improving patient experience, optimizing medical processes, and facilitating data-driven decision-making, and analyzes the current technical barriers and obstacles. The article envisions the future development of LLMs in smart inhalers, advocating for multidisciplinary collaboration to fully harness their potential while effectively addressing associated risks, thereby advancing medical services toward greater intelligence, personalization, and efficiency.

 

MissingLinkText

Keywords

large language models - smart inhalers - patient education - data privacy - medical decision-making

Introduction

Respiration is a continuous and essential physiological activity in daily life and respiratory diseases place a huge burden on patients.[1] This is particularly true for patients with chronic conditions, including asthma and chronic obstructive pulmonary disease (COPD), which are characterized by long treatment cycles, affect the airways, and limit daily activity capabilities.

Inhalation is widely regarded as the optimal route of administration for first-line treatment of asthma and COPD. Inhaled drugs can be delivered directly to pulmonary lesions and are rapidly absorbed due to the unique physiological properties of the lungs, including their large surface area, thin alveolar walls, and high blood flow.[2] The drug can be delivered via pressurized metered dose inhalers,[3] dry powder inhalers,[4] or nebulizers.[5] Inhalation can reduce the relative dosage of medications and minimize the incidence of systemic side effects when compared with oral and injectable routes for pulmonary diseases.

Inhalation parameters, especially inhalation flow rate, inhalation volume, and breath-holding time, contribute to the success of inhaled drug deposition. Therefore, patient adherence to treatment regimens and the correct use of inhalation devices are key factors influencing the distribution of medications within the lungs.[6] However, the inability of patients to use medications correctly and/or nonadherence to treatment plans has been recognized as a problem for decades.[7] [8] Furthermore, pathophysiological characteristics and disease progression are significantly different from patient to patient, and specialized clinicians are often blind to patients' usage patterns and physical conditions during self-medication under traditional outpatient models.[9] This makes it challenging to achieve fully individualized and dynamically optimized inhalation therapy. Therefore, the development of effective adherence management tools and strategies is a crucial direction to ensure that patients are able to use these inhalation devices correctly.[10]

The market has seen a gradual emergence of smart inhalers ([Table 1])[11] [12] [13] [14] [15] [16] [17] targeting chronic respiratory diseases, with adherence to inhaled treatments being an issue.[18] These smart devices utilize the Internet of Medical Things[19] technology and are integrated with smartphone applications to provide health care monitoring without manual intervention. They record patients' medication times and dosages, offer reminders and feedback to help patients take medications on time and in the correct dosage,[20] and monitor patients' inhalation flow rates and patterns, providing real-time feedback to help improve inhalation techniques.[21] [22] Effective clinical assessments can provide personalized treatment methods based on different patients' conditions and responses, help patients evaluate whether their disease is worsening or improving, and, to some extent, enhance patients' adherence and self-management capabilities. However, traditional smart inhalers do not provide clinical assessments. Studies have shown that elderly patients may face difficulties in operating and understanding these devices,[23] and that smart inhalers and their associated applications require certain operational skills, adding a new challenge for some elderly patients.

Table 1 Comparison of marketed smart inhalers

Device name	Release year	Monitoring functions	Type	Brand
ProAir Digihaler	2019	Inhalation airflow, medication time, medication frequency	Built-in smart inhaler	Teva[11]
Propeller Sensor	2014	Medication time, medication frequency, and inhaler remaining dose (varies by model)	External sensor	Propeller Health[12]
Hailie	2017	Inhalation time, medication frequency	External or integrated with an inhaler (varies by model)	Adherium[13]
HeroTracker	2017	Medication frequency, medication time, inhalation airflow (varies by model)	External sensor	Aptar[14]
CapMedic	2022	Inhalation airflow	External smart accessory	Cognita Labs[15]
		Grip angle
FindAir One	2018	Medication time, environmental information	External smart sensor	FindAir[16]
Respiro	2016	Medication operation	External sensor	Amiko[17]

Artificial intelligence (AI) has rapidly developed in recent years. A simple PubMed search using the term “large language models” yielded 2,002 results in 2023 and 4,186 results in 2024, indicating that Large Language Models (LLMs) are an emerging field in AI applications with explosive momentum. LLMs have made significant progress in image recognition and natural language processing (NLP).[24] In the health care industry, LLMs can be used to summarize medical records, recommend clinical pathways, answer medical queries, and assist clinical decision-making.[25] LLMs possess higher semantic understanding capabilities, contextual inference abilities, and the potential to integrate multisource information compared with traditional machine learning (ML) or rule-based systems.[26] In addition, by combining sentiment analysis and natural language understanding capabilities, LLMs can serve as intelligent assistants, providing real-time emotional support and behavioral interventions to further improve patient adherence.

Although LLMs have shown broad application prospects in various health care fields, their specific applications in inhaled drug delivery are still in the exploratory stage. This paper aims to explore the potential of LLMs in enhancing patient adherence in inhaled drug delivery. It analyzes the advantages and challenges and proposes future research directions to promote the integration of inhalation pharmaceutics and LLM technologies, providing a reference for the innovative development of smart medical devices.

 

MissingLinkText

Large Language Models Explores the Potential of Patients Using Digital Inhalers

Current Status of Smart Inhalers

In order to explore how LLMs can enhance the use of smart inhalers in the patient experience, it is first necessary to clarify that the primary goal of smart inhalers is to address medication adherence in the treatment of chronic respiratory diseases. Nonadherence includes initiation failures, operational errors (including inhalation angle and technique), and discontinuation. Globally, nonadherence to chronic disease treatments leads to poor clinical outcomes, reduced quality of life, and high health care and social costs.[27]

Smart inhalers enable real-time monitoring of patients' medication behaviors by integrating sensors, data recording, and transmission technologies. These devices not only record patients' inhalation counts, dosages, and inhalation rate but also use wireless technology to transmit data to smartphones or cloud platforms[28] for viewing and analyzing by patients and health care providers. Additionally, smart inhalers can generate targeted suggestions based on patients' medication data, such as adjusting medication times or dosages to optimize treatment effects, thereby providing personalized medication guidance and health education content. Dierick et al demonstrated through the OUTERSPACE trial that personalized inhalation education using smart inhalers is feasible in adult asthma patients and significantly reduces inhaler operational errors in the short term.[29] From physicians' or professional pharmacists' perspective, smart inhalers provide detailed patient medication data, allowing them to stay informed about patients' medication use[30] and promptly identify improper use or poor adherence issues at all times. Based on these data, doctors can make personalized treatment adjustments, such as modifying drug dosages, changing treatment plans, or scheduling follow-ups. In addition, the health care teams can analyze the data to identify common issues in patient groups and optimize overall treatment strategies.

However, there are numerous challenges in the promotion and application of smart inhalers. First, data privacy and security are significant concerns. Smart inhalers require the collection and transmission of a large amount of patient health data, which involves data privacy and security protection.[31] Therefore, ensuring patient privacy while guaranteeing secure data transmission and storage is a major challenge in the widespread adoption of smart inhalers. Second, technology acceptance is also a key issue. Despite the significant advantages of smart inhalers, some patients may be reluctant to use them due to unfamiliarity with or distrust of new technologies. In practical applications, clinicians and pharmacists need to spend more time explaining how to use smart inhalers correctly,[28] which increases the workload of health care professionals. Therefore, improving the user-friendliness of the devices and providing comprehensive user education are crucial for the successful application of smart inhalers.

 

MissingLinkText

Large Language Models Improve Medication Adherence and Feedback Mechanisms

The core of LLMs lies in their deep neural networks based on the Transformer architecture trained on large-scale text data using self-supervised learning methods.[32] This training approach allows the models to capture grammatical, semantic, and contextual information from the language, enabling them to understand and generate natural language ([Table 2]).[33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] These models, through their multilayer neural network architectures, are capable of handling complex language patterns and exhibit outstanding transfer learning capabilities, which can adapt to specific tasks through fine-tuning to function effectively across various application scenarios. They can not only help patients better understand and manage their conditions but also reduce suboptimal treatment outcomes or adverse reactions due to inappropriate use.[50]

Table 2 Comparison of common Large Language Models

Model name	Capabilities	Open source	Interactivity	References
GPT-o1 (OpenAI)	High-quality text generation and understanding, multimodal input (supports images), comparable to human experts	No	Highly interactive, supports multiturn dialogues, and task automation	[33]
PaLM 2 (Google)	Multilingual understanding and generation, specialized text processing (legal, medical), code generation, and debugging support	No	Highly interactive, suitable for dialogue systems and automation	[34]
Gemini (Google DeepMind)	Comprehensive natural language processing and multimodal capabilities support text and image generation, understanding, and interaction, integrated into the Google ecosystem	No	Highly interactive, supports multiturn dialogues, and task automation	[35]
LLaMA 3 (Meta)	Efficient text generation and understanding, multilingual and cross-domain applications, support customization and fine-tuning	Partially open source	Highly interactive, suitable for research and commercial applications	[36]
Claude 2 (Anthropic)	Highly controllable output, reduces harmful content generation, supports complex tasks, and multiturn dialogues	No	Highly interactive, focused on secure dialogue systems	[37]
BLOOM (Hugging Face)	Multilingual text generation and understanding support 46 languages, community-driven, continuous optimization, and expansion	Yes	Highly interactive, suitable for research and commercial use	[38]
GPT-NeoX-20B (EleutherAI)	Free text generation and understanding support various NLP tasks, including Q&A, summarization, and translation	Yes	Highly interactive, suitable for research and development	[39]
Deepseek-V3 (deepseek-AI)	High-quality text generation and understanding support multiple languages and complex tasks such as code generation, academic writing assistance	No	Highly interactive, suitable for enterprises and developers	[40]
StableLM (Stability AI)	Free-text generation and understanding support various NLP tasks, including dialogue, summarization, and translation	Yes	Highly interactive, suitable for research and development	[41]
M6 (Alibaba DAMO)	Multimodal text and image generation, efficient cross-domain task processing, supports multilingual and multimodal applications	No	Highly interactive, suitable for enterprise-level applications	[42]
PanGu-Σ (Huawei)	High-quality Chinese text generation and understanding support various Chinese NLP tasks, including dialogue, summarization, and translation	No	Highly interactive, suitable for Chinese dialogue systems	[43]
Megatron-Turing NLG (NVIDIA)	An extremely large-scale language model, high-quality text generation, and complex task handling, optimized for deployment efficiency in the NVIDIA hardware and software ecosystem	Partially open source	Highly interactive, suitable for enterprises and research institutions	[44]
Gopher (DeepMind)	Scientific and professional domain text generation and understanding support complex knowledge-intensive tasks, strong information retrieval, and knowledge reasoning abilities	No	Highly interactive, suitable for research and academic applications	[45]
CodeGen (Salesforce)	Multilingual code generation and understanding support code completion, error detection, and repair, suitable for software development assistance tools	Yes	Highly interactive, suitable for development tools and IDEs	[46]
ERNIE 4.0 (Baidu)	Multimodal data processing (text, images, audio) supports multilingual and cross-domain NLP tasks, strong context understanding, and generation capabilities	Partially open source	Highly interactive, suitable for multimodal application scenarios	[47]
GLM (BAAI)	Multilingual text generation and understanding, especially excellent in Chinese NLP tasks, supports multitask learning and low-resource languages	Partially open source	Highly interactive, suitable for research and commercial applications	[48]
Qwen (Zhipu.AI)	High-quality text generation and understanding support multilingual and multimodal tasks, especially optimized for Chinese processing, and supports dialogue systems, text summarization, translation, Q&A, etc.	Partially open source	Highly interactive, suitable for customer service, virtual assistants, etc.	[49]

Abbreviations: IDE, integrated development environment; LLMs, large language models; NLP, natural language processing.

Sezgin et al investigated the clinical accuracy of LLMs in answering questions related to postpartum depression.[51] The results demonstrated that LLMs were able to answer with high accuracy and clinical relevance, as well as consistency in clinically pertinent answers, which is crucial for health care applications. This finding opens up the possibility of using LLMs as virtual health assistants for smart inhalers. LLMs can interact with patients through natural language and use data reasoning to provide detailed, personalized medication guidance and health education, significantly enhancing the accuracy and efficiency of patient consultations and medical education. Furthermore, LLMs can solve various questions encountered by patients during medication usage in real-time, thereby increasing patients' sense of control and trust in their treatment process.

Many patients with chronic respiratory diseases fail to adhere strictly to prescribed medication regimens in self-management for various reasons. Existing smart inhalers improve medication adherence by providing medication reminders and monitoring breathing. LLMs can further personalize this aspect of smart inhalers' functionality. Patients exhibit different pathophysiological characteristics, medication responses, and lifestyle habits; these individual difference presents a major challenge in the treatment of chronic respiratory diseases.[52] However, LLMs are able to integrate and learn from patients' medical histories, medication records, real-time monitoring data, and other multisource information. LLMs can also translate patients' lay language and self-reported information into data.[53] [54] Based on in-depth analysis and understanding of patient data, and utilizing appropriate prompt templates,[55] LLMs can provide specific medication recommendations, such as adjusting medication timing, dosage, or changing medication types, to accommodate patients' dynamic health conditions ([Fig. 1]).

Fig. 1 Feedback flowchart for LLMs in personalized inhalation treatment plans. LLMs, large language models.

LLMs themselves are primarily trained on text data and do not have inherent image generation capabilities; however, through modern intelligent medical systems, they can be integrated with specialized image generation or computer vision models to achieve multimodal tasks. Javan et al utilized tools such as Midjourney and DALL-E 3 to visualize symptoms like dizziness or tinnitus, enhancing understanding between patients and providers.[56] Patients supplement their primary symptoms by selecting images and using AI-generated differential diagnoses. It is conceivable to apply this technology to digital inhalers. When the system detects that the patient's inhalation angle is incorrect, it can invoke a visual model to generate a schematic diagram and provide real-time feedback on the correct usage posture through a web interface or an App. Combined with the text explanations generated by the LLMs, it will offer integrated “image + text” personalized interactive guidance. This multimodel collaborative approach takes advantage of the strengths of LLMs in natural language understanding and generation, and the illustrative capabilities of visual models, to comprehensively enhance the user experience and educational effectiveness of digital inhalers.

In addition, it is foreseeable that LLMs can analyze patients' medication records to identify irregular medication use or missed doses, alerting patients to the long-term risks of such behaviors and suggesting adjustments to their medication plans or further health assessments. This dynamic feedback mechanism not only improves medication adherence but also helps medical teams understand patients' health conditions more promptly for corresponding treatment adjustments.[57] The advent of LLMs has also significantly reduced the workload of health care workers and improved the efficiency and quality of personalized medicine.

The application of LLMs in digital inhalers can also enhance patients' active participation and overall health management in various ways. This is thanks to the monitoring functions of digital inhalers. Excitingly, some technologies can correlate environmental factors that may trigger asthma (such as allergens and pollutants) with the patient's geographical location. When patients are reexposed to the same set of triggers in the future, the digital inhaler can alert them to potential risks in advance.[58] LLMs can also dynamically adjust medication plans based on real-time monitoring data such as patients' medication habits, inhalation flow rates, and environmental factors. For example, LLMs might summarize patterns in patients' medication timing and locations, suggest more reasonable medication schedules, and reiterate the potential adverse consequences of nonadherence, encouraging patients to take their medication voluntarily. Beyond passively responding to patients' queries, LLMs can also proactively ask patients questions based on reinforcement learning through preset dialogue strategies to understand their current physical sensations and discomforts, thereby adjusting medication recommendations. Unlike traditional methods that rely on external prompts, the system can automatically trigger exemplary dialogues based on patients' medication records and real-time physiological data, such as: “after using the inhaler just now, do you feel chest tightness or increased phlegm?” or “Has the frequency of nighttime wheezing increased in the past 2 days?” Through such proactive inquiries, the system can more comprehensively collect patients' subjective feedback, ensuring they are in the optimal treatment state and avoiding information omissions or misunderstandings that may arise from relying solely on fixed prompts.

 

MissingLinkText

Multilingual Support and Cultural Adaptation

In a diverse medical environment, language and cultural differences become barriers to patients accessing health information and following treatment plans. However, LLMs have powerful multilingual processing capabilities[59] to provide accurate medication guidance and health education for patients from different language backgrounds. For example, LLMs can generate instructions for use and health recommendations for smart inhalers in the patient's native language or convey information through role-playing and language simulation at patient requests, ensuring accurate understanding of information and reducing medication errors and suboptimal treatment outcomes due to language barriers. In addition, LLMs demonstrate the ability to recognize emotion,[60] adjusting the expression and content of health recommendations based on the contextual information from a dialogue with patients, thereby enhancing patients' acceptance and adherence to treatment and improving the inclusivity and effectiveness of device usage.

 

MissingLinkText

 

MissingLinkText

Clinical Decision Support Systems in Smart Inhalers Using Large Language Models

Today, most smart inhalers and devices can collect vast amounts of medical data. LLMs, with their powerful NLP, image recognition, and deep data analysis capabilities, show significant potential in providing clinical decision support for smart inhalers.[61] Clinical Decision Support Systems encode clinical knowledge into computerized algorithms and integrate them with patient-specific data to provide information and decision guidance to clinicians.[62] This operational approach aligns with data transformation and analysis capabilities of LLMs, which excel in their ability to perform real-time computations, provide timely health assessments and treatment recommendations, and predict the outcomes of user actions and decisions. Furthermore, in a complex medical environment, LLMs can integrate data from electronic medical record systems[63] and remote monitoring devices,[64] offering comprehensive health management plans and explaining their decision rationales, enabling clinicians to make more informed decisions.[48] [65] This decision-making process is not one-directional to patients; LLMs can also help health care professionals better understand patients' health conditions, thereby developing more precise and efficient treatment plans ([Fig. 2]).

Fig. 2 Simulation workflow diagram for LLMs in clinical decision support systems. LLMs, large language models.

For certain asthma patients, LLMs may, based on the patient's latest inhalation flow data, medication adherence records, and relevant clinical research evidence, recommend increasing the use of anti-inflammatory medications to control inflammatory responses, thereby improving the patient's condition. These decision support functions not only enhance the scientific rigor and efficiency of medical decisions but also promote collaboration between different health care systems, optimizing the overall medical service process.

 

MissingLinkText

Challenges and Future Directions

Despite the significant potential of LLMs in empowering smart inhalers, their practical application still faces numerous challenges.[66] [67]

Data Privacy and Security

In Smart Inhalers, data must be uploaded to the cloud. Data security and privacy protection are key considerations. LLMs need to handle vast amounts of sensitive patient health data, including medication records, respiratory parameters, and personal health information. Ensuring the privacy and security of these data is paramount and is one of the main barriers to successful clinical application of AI.[68]

The General Data Protection Regulation, the Health Insurance Portability and Accountability Act, and other privacy protection regulations that exist impose strict requirements on the processing of medical data. When processing such data, LLMs must ensure encrypted transmission and storage to prevent unauthorized access and data breaches. Existing data encryption methods may address the security concerns of LLMs when processing patient medical data. Sensitive medical data can be encrypted using homomorphic encryption algorithms. Homomorphic encryption allows computations to be performed directly on encrypted data without requiring decryption.[69] This allows LLMs to analyze encrypted data without accessing the plaintext. The final results are decrypted only when necessary by authorized parties (e.g., patients or doctors) who possess the private key. This approach ensures data security during both transmission and processing.

Other studies have proposed blockchain technology as an alternative solution to prevent data breaches in LLMs.[70] [71] [72] Blockchain uses distributed ledger technology to guarantee data immutability and traceability. Each data block contains the hash of the previous block, forming an unalterable chain. Kumar et al proposed a privacy-preserving method that combines blockchain and homomorphic encryption for medical image analysis.[69] This technique safeguards data privacy through homomorphic encryption while using blockchain to ensure the traceability of model updates. Similarly, Tan' group developed a blockchain-based platform for secure data sharing in ophthalmic research.[73] In this system, patient data are encrypted using homomorphic encryption and uploaded to the cloud. LLMs generate recommendations, and the results are logged on the blockchain.

 

MissingLinkText

Model Accuracy and Reliability

Although LLMs have been applied and explored in many fields, their accuracy and reliability remain key concerns in the stringent medical domain.[24] [74] [75] LLMs may exhibit “hallucinations,” generating inaccurate or misleading responses that could “deceptively persuade” patients. The application of LLMs requires high levels of accuracy and reliability, in smart inhalers especially in clinical decision support. Currently, the U.S. Food and Drug Administration (FDA) has not approved LLM as a Clinical Decision Support devices.[76]

GPT-4 is OpenAI's latest model. It incorporates chain-of-thought reasoning, enabling users to handle more complex medical queries. This model has advanced reasoning capabilities to reduce hallucinations and resolve access discrepancies,[77] suggesting that changes in the thinking patterns of LLMs can enhance their applicability in the medical field. In addition, retraining LLMs with data specifically tailored to the inhalation medical field can improve the model's understanding of specialized terminology and medical knowledge. Clinical trials and feedback from actual applications are essential to evaluate the reliability and safety of LLMs in clinical settings. A combination of expert systems and rule engines to review and correct LLM outputs ensures compliance with medical standards. Continuous optimization of model performance and the incorporation of high-quality data are foreseeable methods to improve the reliability of model outputs. In addition, medical incidents due to LLMs' hallucinations and erroneous recommendations require clear accountability, and an appropriate legal framework and compensation mechanisms to regulate the application of LLMs in the medical field.

 

MissingLinkText

User Acceptance and Experience

Although LLMs can provide personalized medication guidance and health advice, patients may develop a certain amount of doubts and distrust towards AI technologies.[78] To improve user acceptance, human supervision is likely to remain indispensable in the future, as continuously training and optimizing the interactive methods and content of LLMs alone cannot fully alleviate patients' concerns. In addition, it is important to conduct educational and training activities to help patients become familiar with and proficient in using LLM-assisted smart inhalers to improve the user experience.

 

MissingLinkText

Future Outlook

Looking ahead, as NLP and ML technologies continue to advance, the accuracy and reliability of LLMs will further improve, enabling better support for medical decision-making and patient management. Furthermore, interdisciplinary collaboration will promote the integration of LLMs with other advanced technologies, such as the Internet of Things, blockchain, and virtual reality, thereby creating a more intelligent and comprehensive health care ecosystem. Additionally, as global medical data standards gradually unify and mature, LLMs will achieve broader application worldwide, facilitating the sharing and optimization of medical resources and driving the development of smart inhalers toward greater intelligence, personalization, and efficiency.

 

MissingLinkText

 

MissingLinkText

Conclusion

The application of LLMs in smart inhalers shows great potential in significantly enhancing the intelligence of the devices, optimizing patient experience, and promoting the precision of medical decision-making. LLMs may also lead to lower costs by effectively reducing the manual labor involved in personalized treatment. Through NLP and deep data analysis, LLMs can provide patients with personalized medication guidance and health recommendations while assisting health care professionals in more efficiently managing and interpreting patient data, thereby optimizing treatment plans. However, to realize these advantages, it is essential to first overcome challenges related to data privacy and security, model accuracy and reliability, and establish robust regulatory and protection mechanisms. Only then can LLMs truly fulfill their transformative role in the medical field. Similar to other digital health care tools such as telemedicine and image-based deep learning algorithms, LLMs application requires rigorous evaluation and governance to ensure that they enhance the quality of medical services without introducing new ethical and legal issues.[79]

In summary, LLMs offer new technical possibilities and application prospects for the intelligent upgrading of smart inhalers in the field of inhalation therapy. Through continuous research and development, combined with stringent regulatory and ethical standards, LLMs will play a crucial role in improving the management of chronic respiratory diseases, enhancing patients' quality of life, and advancing medical services toward greater intelligence, personalization, and efficiency.

 

MissingLinkText

 

MissingLinkText

Conflict of Interest

None declared.

• References

• 1 Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med 2006; 3 (11) e442
• 2 Labiris NR, Dolovich MB. Pulmonary drug delivery. Part I: physiological factors affecting therapeutic effectiveness of aerosolized medications. Br J Clin Pharmacol 2003; 56 (06) 588-599
• 3 Vallorz E, Sheth P, Myrdal P. Pressurized metered dose inhaler technology: manufacturing. AAPS PharmSciTech 2019; 20 (05) 177
• 4 Hickey AJ. Dry powder inhalers: an overview. J Aerosol Med Pulm Drug Deliv 2023; 36 (06) 316-323
• 5 Martin AR, Finlay WH. Nebulizers for drug delivery to the lungs. Expert Opin Drug Deliv 2015; 12 (06) 889-900
• 6 Newman SP. Drug delivery to the lungs: challenges and opportunities. Ther Deliv 2017; 8 (08) 647-661
• 7 Haughney J, Price D, Kaplan A. et al. Achieving asthma control in practice: understanding the reasons for poor control. Respir Med 2008; 102 (12) 1681-1693
• 8 Melani AS, Bonavia M, Cilenti V. et al; Gruppo Educazionale Associazione Italiana Pneumologi Ospedalieri. Inhaler mishandling remains common in real life and is associated with reduced disease control. Respir Med 2011; 105 (06) 930-938
• 9 Pritchard JN, Nicholls C. Emerging technologies for electronic monitoring of adherence, inhaler competence, and true adherence. J Aerosol Med Pulm Drug Deliv 2015; 28 (02) 69-81
• 10 Lavorini F. Inhaled drug delivery in the hands of the patient. J Aerosol Med Pulm Drug Deliv 2014; 27 (06) 414-418
• 11 Teva. Teva launches two digital inhalers in the U.S., AirDuo Digihaler (fluticasone propionate and salmeterol) inhalation powder and ArmonAir Digihaler (fluticasone propionate) inhalation powder. Accessed January 21, 2025 at: https://www.tevapharm.com/news-and-media/latest-news/teva-launches-two-digital-inhalers-in-the-u.s.-airduo-digihaler-fluticasone-propionate-and-salmeterol/
• 12 Propeller Health. Get back to doing the things you love. Accessed January 21, 2025 at: https://propellerhealth.com/
• 13 Adherium. Adherium Limited is an international respiratory ehealth company focused on patient adherence, remote monitoring and data management solutions for patients, physicians, payers, and providers. Accessed January 21, 2025 at: https://adherium.com/
• 14 Aptar. Aptar Pharma Launches HeroTracker Sense. Accessed January 21, 2025 at: https://aptar.com/news-events/aptar-pharma-launches-herotracker-sense-smart-connected-healthcare-device/
• 15 Cognita Labs. Your personal inhaler guide that goes wherever you do!. Accessed January 21, 2025 at: https://capmedicinhaler.com/product/?srsltid=AfmBOooNUNSpMjcdvURb6XK-4TTCT0JLLDH2Jc_9NGsPcxCQmrpvJHEh
• 16 FindAir. Smart inhalers that make remote asthma & COPD care a reality. Accessed January 21, 2025 at: https://findair.eu/
• 17 Amiko. Upgrading respiratory care with digital medicines. Accessed January 21, 2025 at: https://amiko.io/
• 18 Häußermann S, Arendsen LJ, Pritchard JN. Smart dry powder inhalers and intelligent adherence management. Adv Drug Deliv Rev 2022; 191: 114580
• 19 Manickam P, Mariappan SA, Murugesan SM. et al. Artificial intelligence (AI) and internet of medical things (IoMT) assisted biomedical systems for intelligent healthcare. Biosensors (Basel) 2022; 12 (08) 562
• 20 Sulaiman I, Greene G, MacHale E. et al. A randomised clinical trial of feedback on inhaler adherence and technique in patients with severe uncontrolled asthma. Eur Respir J 2018; 51 (01) 1701126
• 21 Taylor TE, Holmes MS, Sulaiman I, Costello RW, Reilly RB. Monitoring inhaler inhalations using an acoustic sensor proximal to inhaler devices. J Aerosol Med Pulm Drug Deliv 2016; 29 (05) 439-446
• 22 Braido F, Paa F, Ponti L. et al. A new tool for inhalers' use and adherence monitoring: the Amiko validation trial. Int J Eng Res Sci 2016; 2: 159-166
• 23 Pothirat C, Chaiwong W, Phetsuk N, Pisalthanapuna S, Chetsadaphan N, Choomuang W. Evaluating inhaler use technique in COPD patients. Int J Chron Obstruct Pulmon Dis 2015; 10: 1291-1298
• 24 Pais C, Liu J, Voigt R, Gupta V, Wade E, Bayati M. Large language models for preventing medication direction errors in online pharmacies. Nat Med 2024; 30 (06) 1574-1582
• 25 Yang JM, Chen BJ, Li RY. et al. Artificial intelligence in medical metaverse: applications, challenges, and future prospects. Curr Med Sci 2024; 44 (06) 1113-1122
• 26 Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019; 25 (01) 44-56
• 27 Tommelein E, Mehuys E, Van Hees T. et al. Effectiveness of pharmaceutical care for patients with chronic obstructive pulmonary disease (PHARMACOP): a randomized controlled trial. Br J Clin Pharmacol 2014; 77 (05) 756-766
• 28 Hui CY, McKinstry B, Mclean S, Buchner M, Pinnock H. Assessing the technical feasibility of a flexible, integrated Internet-of-things connected for asthma (C4A) system to support self-management: a mixed method study exploring patients and healthcare professionals perspectives. JAMIA Open 2022; 5 (04) ooac110
• 29 Dierick BJH, Achterbosch M, Eikholt AA. et al. Electronic monitoring with a digital smart spacer to support personalized inhaler use education in patients with asthma: the randomized controlled OUTERSPACE trial. Respir Med 2023; 218: 107376
• 30 Liau JC, Ho CY. Intelligence IoT (Internal of Things) telemedicine health care space system for the elderly living alone. Paper presented at: 2019 IEEE Eurasia conference on biomedical engineering, healthcare and sustainability (ECBIOS); May 31–June 3, 2019; Okinawa, Japan
• 31 Attaway AH, Alshabani K, Bender B, Hatipoğlu US. The utility of electronic inhaler monitoring in COPD management: promises and challenges. Chest 2020; 157 (06) 1466-1477
• 32 Zhang Y, Liu C, Liu M. et al. Attention is all you need: utilizing attention in AI-enabled drug discovery. Brief Bioinform 2023; 25 (01) bbad467
• 33 Han W, Wan C, Shan R. et al. Evaluation of error detection and treatment recommendations in nucleic acid test reports using ChatGPT models. Clin Chem Lab Med 2025;
• 34 Singhal K, Azizi S, Tu T. et al. Large language models encode clinical knowledge. Nature 2023; 620 (7972) 172-180
• 35 Temsah A, Alhasan K, Altamimi I. et al. DeepSeek in healthcare: revealing opportunities and steering challenges of a new open-source artificial intelligence frontier. Cureus 2025; 17 (02) e79221
• 36 Adams LC, Truhn D, Busch F. et al. Llama 3 challenges proprietary state-of-the-art large language models in radiology board-style examination questions. Radiology 2024; 312 (02) e241191
• 37 Agarwal M, Goswami A, Sharma P. Evaluating ChatGPT-3.5 and Claude-2 in answering and explaining conceptual medical physiology multiple-choice questions. Cureus 2023; 15 (09) e46222
• 38 Hugging Face. BigScience Large Open-science Open-access Multilingual Language Model. . Accessed April 20, 2025 at: https://huggingface.co/bigscience/bloom
• 39 Hugging Face. EleutherAI/GPT-NeoX-20B. . Accessed April 20, 2025 at: https://huggingface.co/EleutherAI/gpt-neox-20b
• 40 Hugging Face. DeepSeek-V3–0324. . Accessed April 20, 2025 at: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324
• 41 Hugging Face. Stabilityai/StableLM-2–1_6b. . Accessed April 20, 2025 at: https://huggingface.co/EleutherAI/gpt-neox-20b
• 42 Towards Data Science. Meet M6–10 trillion parameters at 1% GPT-3′s energy cost. . Accessed April 20, 2025 at: https://towardsdatascience.com/meet-m6-10-trillion-parameters-at-1-gpt-3s-energy-cost-997092cbe5e8/
• 43 Huawei Cloud. PanGu. Accessed April 20, 2025 at: https://www.huaweicloud.com/product/pangu.html
• 44 Nvidia Developer. Using DeepSpeed and Megatron to train Megatron-Turing NLG 530B, the World's Largest and Most Powerful Generative Language Model. Accessed April 20, 2025 at: https://developer.nvidia.com/blog/using-deepspeed-and-megatron-to-train-megatron-turing-nlg-530b-the-worlds-largest-and-most-powerful-generative-language-model/
• 45 DeepMind. Language modelling at scale: Gopher, ethical considerations, and retrieval. Accessed April 20, 2025 at: https://deepmind.google/discover/blog/language-modelling-at-scale-gopher-ethical-considerations-and-retrieval
• 46 Nijkamp E, Pang B, Hayashi H. et al. CodeGen: an open large language model for code with multi-turn program synthesis. arXiv. Preprint. February 27, 2023. Accessed at: https://doi.org/10.48550/arXiv.2203.13474
• 47 AI Mode. What is ERNIE 4.0?. Accessed April 20, 2025 at: https://aimode.co/model/ernie-4/
• 48 Hugging Face. THUDM/GLM-10b-chinese. . Accessed April 20, 2025 at: https://huggingface.co/THUDM/glm-10b-chinese
• 49 Qwen. Qwen 2.5-Max: Free AI Chat. Accessed April 20, 2025 at: https://qwen.org/
• 50 Blalock SJ, Solow EB, Reyna VF. et al. Enhancing patient understanding of medication risks and benefits. Arthritis Care Res (Hoboken) 2022; 74 (01) 142-150
• 51 Sezgin E, Chekeni F, Lee J, Keim S. Clinical accuracy of large language models and Google search responses to postpartum depression questions: cross-sectional study. J Med Internet Res 2023; 25: e49240
• 52 Wang X. New strategies of clinical precision medicine. Clin Transl Med 2022; 12 (02) e135
• 53 Flaharty KA, Hu P, Hanchard SL. et al. Evaluating large language models on medical, lay-language, and self-reported descriptions of genetic conditions. Am J Hum Genet 2024; 111 (09) 1819-1833
• 54 He Z, Bhasuran B, Jin Q. et al. Quality of answers of generative large language models versus peer users for interpreting laboratory test results for lay patients: evaluation study. J Med Internet Res 2024; 26: e56655
• 55 Meskó B. Prompt engineering as an important emerging skill for medical professionals: tutorial. J Med Internet Res 2023; 25: e50638
• 56 Javan R, Cole J, Hsiao S, Cronquist B, Monfared A. Integration of AI-generated images in clinical otolaryngology. Cureus 2024; 16 (08) e68313
• 57 Hakizimana A, Devani P, Gaillard EA. Current technological advancement in asthma care. Expert Rev Respir Med 2024; 18 (07) 499-512
• 58 Greene G, Costello RW. Personalizing medicine - could the smart inhaler revolutionize treatment for COPD and asthma patients?. Expert Opin Drug Deliv 2019; 16 (07) 675-677
• 59 Moser D, Bender M, Sariyar M. Generating synthetic healthcare dialogues in emergency medicine using large language models. Stud Health Technol Inform 2024; 321: 235-239
• 60 Rathje S, Mirea DM, Sucholutsky I, Marjieh R, Robertson CE, Van Bavel JJ. GPT is an effective tool for multilingual psychological text analysis. Proc Natl Acad Sci U S A 2024; 121 (34) e2308950121
• 61 Xu Y, Liu X, Cao X. et al. Artificial intelligence: a powerful paradigm for scientific research. Innovation (Camb) 2021; 2 (04) 100179
• 62 Eberhardt J, Bilchik A, Stojadinovic A. Clinical decision support systems: potential with pitfalls. J Surg Oncol 2012; 105 (05) 502-510
• 63 Abid M, Schneider AB. Clinical informatics and the electronic medical record. Surg Clin North Am 2023; 103 (02) 247-258
• 64 Mosnaim GS, Greiwe J, Jariwala SP, Pleasants R, Merchant R. Digital inhalers and remote patient monitoring for asthma. J Allergy Clin Immunol Pract 2022; 10 (10) 2525-2533
• 65 Miller T. Explanation in artificial intelligence: insights from the social sciences. Artif Intell 2019; 267: 1-38
• 66 Xu X, Chen Y, Miao J. Opportunities, challenges, and future directions of large language models, including ChatGPT in medical education: a systematic scoping review. J Educ Eval Health Prof 2024; 21: 6
• 67 Tian S, Jin Q, Yeganova L. et al. Opportunities and challenges for ChatGPT and large language models in biomedicine and health. Brief Bioinform 2024; 25 (10) bbad493
• 68 Khalid N, Qayyum A, Bilal M, Al-Fuqaha A, Qadir J. Privacy-preserving artificial intelligence in healthcare: techniques and applications. Comput Biol Med 2023; 158: 106848
• 69 Kurniawan H, Mambo M. Homomorphic encryption-based federated privacy preservation for deep active learning. Entropy (Basel) 2022; 24 (11) 1545
• 70 Kumar R, Kumar J, Khan AA. et al. Blockchain and homomorphic encryption based privacy-preserving model aggregation for medical images. Comput Med Imaging Graph 2022; 102: 102139
• 71 Betzler BK, Chen H, Cheng CY. et al. Large language models and their impact in ophthalmology. Lancet Digit Health 2023; 5 (12) e917-e924
• 72 Tan TE, Anees A, Chen C. et al. Retinal photograph-based deep learning algorithms for myopia and a blockchain platform to facilitate artificial intelligence medical research: a retrospective multicohort study. Lancet Digit Health 2021; 3 (05) e317-e329
• 73 Ng WY, Tan TE, Xiao Z. et al. Blockchain technology for ophthalmology: coming of age?. Asia Pac J Ophthalmol (Phila) 2021; 10 (04) 343-347
• 74 Lee P, Bubeck S, Petro J. Benefits, limits, and risks of GPT-4 as an AI Chatbot for medicine. N Engl J Med 2023; 388 (13) 1233-1239
• 75 Arvind Barge S, Mary GI. Improving dependability with low power fault detection model for skinny-hash. PLoS One 2024; 19 (12) e0316012
• 76 Weissman G, Mankowitz T, Kanter G. Large language model non-compliance with FDA guidance for clinical decision support devices. Res Sq 2024; rs.3.rs-4868925
• 77 Temsah MH, Jamal A, Alhasan K, Temsah AA, Malki KH. OpenAI o1-preview vs. ChatGPT in healthcare: a new frontier in medical AI reasoning. Cureus 2024; 16 (10) e70640
• 78 Laux J. Institutionalised distrust and human oversight of artificial intelligence: towards a democratic design of AI governance under the European Union AI Act. AI Soc 2024; 39 (06) 2853-2866
• 79 Shafi S, Parwani AV. Artificial intelligence in diagnostic pathology. Diagn Pathol 2023; 18 (01) 109

Publikationsverlauf

Eingereicht: 24. Januar 2025

Angenommen: 09. Juli 2025

Artikel online veröffentlicht:
18. August 2025