Use of continuous glucose monitoring systems in pediatric patients in the perioperative environment: Challenges and machine learning opportunities

Abstract

Pediatric type 1 diabetes (T1D) is a lifelong condition requiring meticulous glucose management to prevent acute and chronic complications. Conventional management of diabetic patients does not allow for continuous monitoring of glucose trends, and can place patients at risk for hypo- and hyperglycemia. Continuous glucose monitors (CGMs) have emerged as a mainstay for pediatric diabetic care and are continuing to advance treatment by providing real-time blood glucose (BG) data, with trend analysis aided by machine learning (ML) algorithms. These predictive analytics serve to prevent against dangerous BG variations in the perioperative environment for fasted children undergoing surgical stress. Integration of CGM data into electronic health records (EHR) is essential, as it establishes a foundation for future technologic interfaces with artificial intelligence (AI). Challenges in perioperative CGM implementation include equitable device access, protection of patient privacy and data accuracy, ensuring institution of standardized protocols, and financing the cumbersome healthcare costs associated with staff training and technology platforms. This paper advocates for implementation of CGM data into the EHR utilizing multiple facets of AI/ML algorithms.

Key Words: Continuous glucose monitor; Continuous glucose monitoring system; Type 1 diabetes mellitus; Artificial intelligence; Electronic health records

Core Tip: Continuous glucose monitoring systems (CGMS) have become the standard of care for many diabetic patients, offering more precise disease management and reducing the risk of hypo- and hyperglycemia. While CGMs provide continuous data, their accuracy and reliability during periods of surgical stress and anesthetic agent use can be variable and warrants further research to validate their use perioperatively. AI algorithms can analyze CGM trends, detect patterns, and provide personalized insights directly within the electronic health record. This facilitates earlier interventions, improves glycemic control, and supports clinical workflows by reducing data overload.

INTRODUCTION

The incidence of diabetes in the pediatric population is on the rise and the technologic advances in glucose monitoring are proving to be a reliable mainstay treatment. Thus, we are seeing an increase in pediatric patients presenting for surgery with continuous glucose monitors (CGM) and automated insulin dosing (AID) systems controlled by applications on electronic devices, often referred to collectively as CGM systems (CGMS). This new frontier of diabetes management in the operating room is placing anesthesiologists in a new arena of glucose management. We now have access to a patients’ glucose readings without the need for a fingerstick and are able to see average and trending glucose values in a well-constructed manner. We not only need to gain familiarity with these devices but also establish perioperative protocols to streamline glucose monitoring in an efficient and ethical manner. As we advance in the use of CGM, it is crucial to anticipate and address evolving challenges related to data privacy, cybersecurity and an integration with the electronic health record (EHR).

A number of challenges facing CGM utilization include cost and resource limitations, accuracy and reliability in the perioperative environment, healthcare system integration, and ethical and legal considerations. Artificial intelligence (AI) has the potential to greatly enhance CGMS by providing more precise data analysis and personalized therapeutic recommendations. Additionally, it can provide functionality through integration with other devices, enhanced accessibility and big data analysis, thus allowing for early detection and prevention of complications. This manuscript will outline the challenges facing patients and healthcare systems that deal with CGMS, as well as the promising advances that machine learning (ML) has to offer diabetic patients and their perioperative physicians.

BACKGROUND

Whereas capillary blood glucose (BG) monitoring provides static results, CGM technology has the dynamic capability of capturing BG levels over time, with a focus on hypo- and hyperglycemic changes and patterns, allowing clinicians to more accurately gauge glycemic trends and guide therapy. Multiple studies have demonstrated that CGM-derived time in range (TIR, 70–180 mg/dL or 3.9–10.0 mmol/L) levels correlate closely with hemoglobin A1c (HbA1c) values[1]. Beck et al[2] have also found that TIR values can be utilized to extrapolate diabetes-related complications, including nephropathy, retinopathy, neuropathy and cardiovascular complications. Multiple professional societies including the Endocrine Society, American Diabetes Association, and the American Association of Clinical Endocrinology have declared CGMs the standard of care for pediatric patients with type 1 diabetes (T1D)[3-5].

Several CGMS are currently available on the market for use in children (Table 1), featuring rapidly evolving technology and consisting of multiple integral intercommunicating components. A skin sensor transmits glucose values every 5 minutes through a wireless transmitter paired to a Bluetooth capable receiver or smart device with a corresponding application to view and interpret values. The device will record real-time CGM readings at specified time intervals, utilizing glycemic data to generate averages and predictive trends, as well as generate high and low alarms. These glucose readings are sent to the patient’s insulin pump, also referred to as a continuous subcutaneous insulin infusion system, which adjusts insulin dosing using preset algorithms[6,7]. While these wearable devices have not been approved by the Food and Drug Administration (FDA) for use in hospital settings at the time of this manuscript, their transition to hospital use has been under debate and the subject of several major diabetes societies and consensus guidelines. The American Diabetes Association, The Society for Pediatric and Adolescent Diabetes (ISPAD), The Endocrine Society and The Diabetes Technology Society have supported the continued use of CGMs during hospitalization in addition to point-of-care (POC) testing for validation in adults and children[8-10].

Table 1 Currently available continuous glucose monitoring systems used in children in the United States.

Company	Dexcom, Inc	Abbott laboratories	Medtronic	Senseonics
Product	Dexcom G6	FreeStyle Libre 2	Guardian connect	Eversense
Approved age of use (years)	≥ 2	≥ 4	≥ 14	≥ 18
Sensor location	Abdomen	Upper arm	Abdomen or upper arm	Upper arm
Communication technique	Cloud-based wireless system	Bluetooth with smartphone. Uploads to cloud with app	Bluetooth to smartphone. Uploads to cloud	Bluetooth to smartphone. Cloud storage
Year of FDA approval	2018	2020	2018	2018
Special features	Glucose streaming to smartphone, customizable alarms, compatible with smartwatches	14-day wear, real-time alarms. Option for high and low glucose alerts	Predictive alerts (60 minutes before reaching low/high levels, real-time glucose values, predictive alerts, critical alerts customizable alerts)	Long wear sensor (up to 90 days in United States), real-time glucose data, alerts and alarms signaled with on-body vibration, customizable alarms through app
Ability to integrate with insulin pump	Yes	Some	Yes	Yes
Smartphone requirement	Dexcom Clarity	LibreView	Carelink	Eversense data management system pro
Pharmacologic interferences	Acetaminophen, hydroxyurea	Vitamin C	Acetaminophen	Acetaminophen, mannitol or sorbitol
Contraindications	MRI, X ray, diathermy	MRI, X ray, diathermy	MRI, X ray, diathermy	MRI, diathermy, lithotripsy
Device classification	Real-time glucose display	Intermittent scanned	Real-time glucose display	Implant
EHR compatibility potential	Yes	Yes	Yes	Yes

EHR: Electronic health record; MRI: Magnetic resonance imaging; FDA: Food and Drug Administration.

There remain challenges in the ability to facilitate CGM device adoption and data integration into the EHR within the health system. Meaningful use standards paved the way for the incorporation and use of an EHR by financially incentivizing quality, efficiency and protection of patient health information. While the CGM closed loop system continues to evolve and gain mass adoption, the challenge of incorporating wearable device data remains a novel challenge. Some hospital systems have partnered with technology start up organizations to help capture this essential patient data into the EHR by creating interfacing capabilities[11]. The creation of standardized protocols and specific standards in the setting of wireless data transmission, privacy, and coding capabilities will serve as the foundation for network interoperability and hospital billing[11-13].

PERIOPERATIVE CHALLENGES OF CGMS

The perioperative use of personal CGMS in a hospital setting can present several critical concerns. Of primary concern is healthcare provider familiarity to ensure safe and effective usage in a timely manner. Therefore, the development of comprehensive detailed hospital policy and protocol implementation on CGMS is essential to reduce knowledge gaps and optimize safety[14]. Potential issues such as mechanical and software problems, alarm failures as well as human errors can confound clinician use of the CGM[15,16]. A preoperative consultation with the patient’s endocrinologist to construct a well formulated glycemic perioperative management plan would be advised[17]. This would be in addition to standard pediatric diabetic recommendations for any elective surgery (serum electrolytes, ketones, and an assessment of glycemic control (Table 2)[18]. This preoperative assessment should be multidisciplinary involving the surgeon, anesthesiologist and endocrinologist to ensure optimized communication and safety[9].

Table 2 Preprocedural checklist recommendations for type 1 diabetes pediatric surgical patients.

	Preprocedural visit checklist for pediatric T1D with CGMS
	Preop checklist for T1D patient as obtained by endocrinologist:
1	Verify type of insulin pump, CGM device and manufacturer
2	Verify patients’ insulin type and basal rate settings
3	Draw HbA1c and electrolyte levels
4	Verify insulin: Carbohydrate ratio, correction factor
5	Check for interference of operative site with CGMS equipment
6	Ensure patient has new sensor, working battery, and pump supplies on day of surgery

T1D: Type 1 diabetes; CGMS: Continuous glucose monitors systems; HbA1c: Hemoglobin A1c.

Regardless of whether a personal CGMS is used, management of diabetic pediatric patients should strictly adhere to the most current ISPAD consensus guidelines (Table 3). These guidelines were originally created in 2018 and updated in 2022 within a well-established consortium[9]. In the 2022 ISPAD guidelines, Kapellen et al[9] continued to recommend that POC measurements are assessed hourly in the perioperative period extending into the immediate phase of the recovery or when oral nutrition is resumed. This approach serves to ensure that there is a reliable backup system for the CGM as well as detection of erroneous readings, especially in the setting of certain physiologic and pharmacologic scenarios.

Table 3 Summary of International Society for Pediatric and Adolescent Diabetes recommendations.

ISPAD 2022 clinical practice guidelines for management of children with diabetes having surgery
Glycemic goals for surgery
BG should be maintained between 90-180 mg/dL
Prevent perioperative hypoglycemia and DKA
Assessment of pediatric T1D prior to surgery/or anesthesia
Preoperative endocrine consultation: Thorough assessment of glycemia, assess for ketones (blood and urine) formalize perioperative glycemic plan
Preoperative care
If patient expected to receive GA, can be admitted to hospital or same-day clinic
Insulin is required, even in fasting state, to avoid DKA
POC BG should be checked and recorded every hour
Intraoperative care
Monitor POC BG every hour and continue in recovery
CGM can be continued intraoperatively but validated with POC BG levels
Postoperative care
Give short or rapid acting insulin (based upon insulin: Carbohydrate ratio or correction factor)
More frequent CGM/POC BG levels recommended for 24-48 hours after surgery due to surgical stressors
Validate CGM readings after exposure to anesthetic agents with POC BG
Target BG levels in postoperative period between 140-180 mg/dL

ISPAD: International Society for Pediatric and Adolescent Diabetes; CGM: Continuous glucose monitors; BG: Blood glucose; POC: Point-of-care; T1D: Type 1 diabetes; DKA: Diabetic ketoacidosis.

There are a number of studies that examine the limitations of CGMs in surgical settings, predominantly in the adult population. In a 2025 study examining the perioperative care of patients using wearable diabetes devices, Cruz et al[14] reported frequent signal loss and reduced sensor accuracy with use of the DexCom G6 CGM in patients undergoing coronary artery bypass surgery, likely due to electrocautery interference. In another recent publication by Herzig et al[19], the accuracy of Dexcom G6 CGMs is significantly reduced during cardiac surgery involving deep hypothermic circulatory arrest, with sensor accuracy inversely correlated to core body temperature. A similar pilot study by Sugiyama et al[20] examined the accuracy of subcutaneous CGM systems in neurosurgical and cardiac surgery patients, indicating lower glucose levels compared to reference measurements during cardiopulmonary bypass.

PHARMACOLOGIC AND PHYSIOLOGIC INTERFERENCES WITH CGMSS

Various factors can compromise the accuracy of CGM values particularly in the setting of hypoperfusion or hypothermia. These conditions can lead to falsely low readings, thereby necessitating further interpretation and cross verification with POC measurements. Of particular concern is the use of acetaminophen, commonly administered in perioperative clinical setting, which can cause falsely elevate readings for up to eight hours[9,17]. This is believed to occur because the phenolic moiety in acetaminophen produces an electrochemical signal that is mistakenly interpreted as glucose by the glucose biosensor[21]. Similarly, susceptibility to pharmacologic interference have also been seen with the use of aspirin, lisinopril, angiotensin receptor blockers, beta blockers, hydroxyurea, albuterol and Vitamin C (Table 4).

Table 4 Common medication interferences on continuous glucose monitor readings.

Medication	Effect on continuous glucose monitor reading
Acetaminophen	↑
Aspirin	↓
Ascorbic acid (Vitamin C)	↓
Lisinopril	↑ or ↓
Losartan	↑ or ↓
Atenolol	↑ or ↓
Albuterol	↑ or ↓
Hydroxyurea	↑ or ↓

In addition to pharmacologic influences, both mechanical and physiological factors can also play a role in CGM accuracy. Decreased perfusion from hypotension or excess pressure on the sensor may create a phenomenon known as compression artifact, altering glucose readings[9,14]. Furthermore, hypothermia can also alter sensor performance. Conversely, hyperthermia can adversely affect sensor performance and signal transmission, while draining battery life. Artificial warming devices should not be placed in direct contact with the sensor, as diaphoresis can weaken adhesives used to secure the sensor. Future large-scale research is needed to assess the accuracy and reliability of CGMS in the perioperative setting to better delineate the technological challenges that have been reported. Until that time, we will need to rely on experienced clinicians to interpret CGM data in conjunction with conventional BG measurement metrics during the perioperative time period.

CONTRAINDICATIONS AND SAFETY RECOMMENDATIONS OF CGMS

All components of the CGMS are absolutely contraindicated in the MRI environment due to concerns for device dysfunction and magnet incompatibility[10]. In the case of X-ray or fluoroscopy, the CGMS should be out of the field of direct radiation or covered with a lead apron. However, according to most manufacturer guidelines, these devices have not been rigorously tested under these circumstances and therefore are still at risk of device malfunction and damage[17]. Larger prospective studies are needed to collect safety data for the use of these devices in areas with radiation. It is strongly advised to keep CGMS away from extreme heat, as electrocautery or diathermy can result in pump malfunction or potential burns, particularly with a metallic infusion needle. The insulin pump should be removed if in close proximity to the surgical field[17].

PERSONAL DATA PRIVACY RISKS/ETHICAL CONSIDERATIONS

CGMS are integral components of the health-related internet of things (H-IoT), a domain of health monitoring devices designed to improve healthcare management[22]. Challenges associated with H-IoT devices include sharing private health data in a way that is both ethical and reliable. Such devices can render a patient vulnerable to privacy and security breaches. Currently, CGM technology commonly uses Bluetooth low energy, an efficient and convenient method for wireless transfer and analysis of glucose trends onto smart devices. Like other wearable devices, CGMS can be susceptible to security breaches, commonly referred to as medical hijacking, or “medjacking”[16]. This term has been described in incidents where unauthorized users can gain control of another medical device. This has been reported in patients with insulin pumps being altered by someone in close proximity[16]. It is essential to strictly enforce stringent privacy and security measures to protect the successful integration of these devices. Addressing these concerns will set the stage for proper optimization and future utilization.

Although there is considerable potential for ML-enhanced CGM use in the perioperative environment, various ethical considerations exist including patient safety, autonomy, equity, and data governance. Importantly, these systems should enhance, rather than replace clinical decision making, as well as provide sufficient guidance for how to supersede inaccurate algorithmic recommendations. In regards to informed consent and autonomy, patients should be made aware of the utilization of ML tools in their care, including how these devices can confer risk and influence patient care related decisions. Ensuring compliance with data privacy laws such as Health Insurance Portability and Accountability Act (HIPAA) is paramount for the big data generated by ML protocols. Breaches are also possible for data integrated with wireless technology or through a cloud-based infrastructure, warranting advanced cybersecurity measures. In terms of liability concerns, if adverse outcomes occur secondary to critical decisions made from ML algorithms, who is to blame- the hospital system, practitioners, or the software vendor? Finally, lack of appropriate integration with hospital-based computer systems can also increase the risk of provider frustration with devices, and the incidence of medical errors.

LIABILITY RISKS

The use of patient owned CGMS devices in hospital settings present challenges that can potentially lead to errors in glucose management. These inaccuracies can place patients at risk for hypo-or hyperglycemic episodes with associated metabolic derangements. In an anesthetized patient who cannot report symptoms, these clinical adverse events could expose healthcare facilities and staff to medico-legal implications[16]. To minimize these risks, it is advised that these patients are managed under the guidance of an endocrinologist, as routine hospital staff may lack familiarity with this technology. Institutions must carefully weigh the risks and benefits before engaging in the use of personal CGMS. A key consideration will necessitate the administration of detailed patient consent/assent, carefully outlining risks of using the device in the hospital setting.

CLINICAL VALIDITY AND FEASIBILITY

The transition from theoretical ML applications to real-world implementation of CGM’s is nuanced and has been verified by a number of clinical trials. Duan et al[23] conducted a multicenter randomized controlled trial (RCT) to evaluate the effectiveness of CGMs compared to conventional monitoring in improving perioperative outcomes in diabetic patients. Conducted in up to 50 secondary and tertiary hospitals in China, the researchers employed ML techniques to analyze and model intraoperative BG changes in an effort to construct a risk-scoring system for high-risk patients[23]. In a similar vein, the Maastricht Study, an observational population-based cohort study which includes pre-diabetic, diabetic and non-diabetic patients utilized ML models to predict glucose levels using prior CGM and accelerometry data. The group concluded that ML models were able to safely and accurately predict glucose values at 15- and 60-minute intervals based on CGM data alone, urging the need for future research in ML implementation for closed-loop insulin delivery systems[24]. Lastly, Shao et al[25] utilized a deep learning model based upon the long short-term memory network to show enhanced performance in hypoglycemia prediction as compared to traditional ML in patients with T1D, suggesting potential for integration with CGM devices to reduce false alarms and provide early warning signs of low BG levels.

There remain a number of regulatory hurdles to FDA approval for ML-enhanced CGM in the perioperative environment, including safety concerns regarding sensor accuracy, device latency, lag time, and liability concerns regarding accountability for perioperative adverse events. The FDA also mandates rigorous ML software validation of performance across various patient demographics and co-existing medical conditions, which is yet incomplete. In addition, significant data gaps continue to exist in terms of accuracy of CGM readings during periods of rapid physiologic change, episodes of hypoperfusion, hypothermia and vasoconstriction. Lastly, robust human factors testing is essential to demonstrate that providers can accurately interpret and respond to device output.

HEALTHCARE DISPARITIES IN CGM UTILIZATION

Access to CGMS in pediatric patients face significant barriers, namely high cost of the device, socioeconomic status (SES), and limited insurance coverage[26,27]. According to the American Diabetes Association^®, populations with the highest prevalence of diabetes - low income, elderly and minority patients - encounter the most challenges in accessing wearable CGM technology. The American Diabetes Association is actively addressing these issues by collaborating with several advocacy groups and healthcare organizations to overcome disparity-related obstacles and optimize diabetic management in these populations[28,29]. Medicaid programs in thirteen states impose stringent eligibility requirements for CGM technology, while the remaining programs lack transparency. For instance, some programs require patients to log four BG fingerstick values per day, yet only cover for three test strips[30]. However, some states have reportedly expanded Medicaid coverage for pediatric T1D patients and demonstrated improved HbA1c values in patients with CGMS[27].

Disparities may exist due to implicit and explicit biases at the provider level. Addala et al[31] found that prescribers, or “gatekeepers” may offer different technologies to pediatric T1D patients with private vs public insurance. Data has shown that insulin-dependent diabetics with Medicaid insurance are 2-5 times less likely to use a CGM, based upon limited accessibility due to public insurance. In addition, despite a higher rate of diabetes in minority populations, Agarwal et al[32] found that amongst 300 young adults stratified by race and ethnicity, fewer Hispanic and non-Hispanic blacks had ever used a CGM compared to their non-Hispanic white counterparts. Therefore, non-Hispanic Black patients had a significantly higher HbA1c level even after accounting for SES. These findings strengthen the assertion that implicit racial biases associated with prescription of standard-of-care diabetes technology have a powerful effect on adverse patient outcomes. A number of steps have been taken to combat the inequities that exist. Medicare and Medicaid programs have increased access by eliminating the 4-times a day minimum BG testing rule in 2021[30]. Additionally, national legislative bodies have mandated private insurance coverage for CGMS. Most importantly, efforts should now be focused on providing CGM technology to publicly-funded insurance programs for low income households and the uninsured.

INTEGRATION OF CGMS TECHNOLOGY INTO HOSPITAL DATA SYSTEMS

The ability to collect, analyze and store data while guaranteeing data integrity and patient privacy is a considerable challenge. In today’s medical landscape, healthcare organizations require the most advanced information technology systems to allow for vital data exchange. At the time of writing of this manuscript, CGM data does not directly integrate with the EHR, leaving providers to rely on screenshots of downloaded reports for documentation in the patient’s chart[33]. It is essential to create standardized protocols with fast and reliable methods to gather clinical data through sophisticated software. These systems require a common language to facilitate interoperability, enabling effective communication between devices and hospital systems. There is no universal approach for the buildout of this specific technology as it would need to be customized, depending heavily on each hospital’s technology infrastructure and policies set forth by the institution. A proposed schematic is depicted in (Figure 1). However, integration of CGMS technology should generally follow the steps as suggested by health level 7 fast healthcare interoperability resource protocols[34] (Table 5).

Figure 1 Schema of personalized continuous glucose monitor integrated into health care system data. BG: Blood glucose; CGM: Continuous glucose monitor; rt glucose: Real-time glucose; HIPPA: Health Insurance Portability and Accountability Act; EHR: Electronic health record; HCP: Healthcare personnel; HL7: Health level 7; API: Application program interface.

Table 5 Summary of the health level 7 fast healthcare interoperability resource protocol criteria.

	HL7 FHIR protocol development criteria
1	Establishing a platform that can connect the technologies between different IT systems
2	Ensure data security for HIPPA regulations while maintaining highest level of patient privacy
3	Adopt or establish protocols which standardize data for ease of interpretation by EHR
4	Application programming interfaces are needed to facilitate communication between CGM cloud-based repository and the hospital EHR
5	Ensure data accuracy with data mapping and integration
6	User interface adjustments-allows personnel to follow data trends and gain insights into patient care specifics
7	Training and support-ongoing education of healthcare providers on how to interpret and use blood glucose data

CGM: Continuous glucose monitors; EHR: Electronic health record; IT: Information technology; HIPPA: Health Insurance Portability and Accountability Act; HL7 FHIR: Health level 7 fast healthcare interoperability resources.

AI AND THE FUTURE OF DIABETIC CARE

The use of AI in the medical field is advancing rapidly, and its application to CGM data management offers considerable promise for improving outcomes in pediatric T1D patients. AI generated insights and prediction analytics would add tremendous value to patient care. While the benefits are clear, the initial buy in would likely be substantial, encompassing data platforms, data storage, hiring of AI experts and staff training. AI implementation would require not only integration into the clinical workflow environment, but also continuous performance monitoring and frequent algorithm updates[35]. These demands could contribute to operational costs and pose several pertinent questions. How would hospital systems absorb additional costs? Would clinicians become less skilled in diabetic management? Can we trust AI generated clinical predictions and suggested therapeutics?

There continues to be mixed data regarding the financial burden or savings potentially associated with the use of CGMs. In regards to the cost of CGMs compared to self-monitoring of BG, a 2022 systematic review of 19 studies by Jiao et al[36] compared the cost-effectiveness of each, finding CGMs to be most cost effective in patients with poor glycemic control or those at risk for severe hypoglycemia. The DIAMOND RCT by Wan et al[37], in contrast, found higher costs associated with CGM device use. This trial compared the cost-effectiveness of initiation of a CGM in patients with T1D compared to multiple daily insulin injections, finding the average total cost over a six-month period for the CGM group to be $11032 compared to $7236 for the self-monitoring group. Of note, while the quality-adjusted life years was similar between the two groups, lifetime projections suggest that a CGM would reduce the risk of chronic diabetic complications.

Several ML applications are designed to work in harmony to produce reliable and efficient predictions for the most effective CGMS management. Each method may employ a combination of algorithms for specific goals of diabetic care[38-40] (Table 6). For example, decision tree learning is an algorithm that falls under the umbrella of supervised learning. Decision trees use categorical variables (i.e. BG values) to label cumulative datasets for the purpose of using historical information to predict outcomes[41]. These AI/ML applications have been found to be particularly useful in four diabetes related domains: Automated retinal screening, predictive population risk stratification, clinical decision support, and patient self-management tools[42,43]. ML has been used to customize interventions for medication adherence and even predict the risk of hospitalization for diabetic patients. Big data analytics have been leveraged to build predictive models to successfully predict long and short-term diabetic complications, including the risk of hypoglycemia[37]. The EHR is a tremendous resource in diabetic management as it can serve as a data repository to develop algorithms for disease prediction, detection, and management[43]. Via decision support systems, AI can give both the patient and their physicians real-time data to customize diabetic treatment plans[44]. The incorporation of CGMS into the hospital setting and the EHR can help shape and advance the future of diabetic care by harnessing the use of AI diagnostic tools.

Table 6 Machine learning applications commonly used in continuous glucose monitors.

AI technique	Predictive modeling	Pattern recognition	Event prediction
Methodology	(1) Supervised learning: Use of labeled data in training models to predict future glucose levels (Hemoglobin A1c); (2) Case based reasoning: Adapts solutions based on previous data history	Unsupervised learning: Use of unlabeled data in training models used to identify patterns and relationships in the data	Supervised learning: Uses datasets with specific outcomes (hypo and hyperglycemia) for event prediction
Algorithms used	(1) Linear regression, decision trees, random forests, neural networks; (2) Similarity measures, nearest neighbor	k-means clustering, isolation forests	support vector machines, logistic regression and deep learning (require more resources for efficiency)
Diabetic applications	Strength: Provides early alerts for high and low glucose readings. Weakness: Requires copious data points for accuracy	Strength: Detects glucose patterns and atypical readings. Weakness: Clustering can require interpretation, can result in false + especially when readings are varied	Strength: Prompts critical alerts in hypoglycemia. Weakness: May produce false positive alarms
Examples in diabetic care	Predictive modeling, self-management tools, retinal screening	Decision support, prediction models, self-management tools, retinal screening	Predictive modeling, patient self-management tools, retinal screening

AI: Artificial intelligence.

AI, while an incredibly powerful tool, is not without significant limitations. As previously discussed, cost, access, implementation and interoperability remain barriers to seamless healthcare system integration. Algorithm bias, seen with differing calibration methods, algorithm designs and data limitations within the CGM platform cannot be overlooked when introducing these devices into the acute setting[45,46]. Devices that have not undergone a sufficient amount of calibration to ensure glycemic accuracy as well as older generations of devices are at higher risk of these complications. Consequences of algorithms bias could cause misinterpretation of glucose values and potentially place patients at risk for acute complications, also putting medical staff in jeopardy of care mismanagement, ultimately causing an erosion of patient confidence in the health care system[47].

It is paramount that these patients are monitored clinically, and continue to have hourly POC measurements until technology can assure the elimination of algorithm bias. A 2024 scoping review by Lim et al[48] examined 22 studies regarding the reliability of CGM accuracy with POC readings in the intraoperative period. The group concluded “high technical reliability” of CGM readings, although inconsistent accuracy compared to conventional monitoring during times of rapid physiological changes. The current research on AI in diabetic care is predicated on retrospective studies, but prospective validation remains to be executed. Hou et al[49] conducted a systematic review of 14 randomized trials for diabetes self-management, concluding that younger patients are more likely to benefit from AI mobile apps, with an enhanced effect seen in tandem with physician feedback. More robust research in this domain is needed to align AI with healthcare needs.

CONCLUSION

CGMS in the perioperative environment have the potential to transform care of children with TID by enhancing glycemic control, thereby minimizing the risk of complications and improving surgical outcomes. The additional application of sophisticated AI algorithms integrated in the EHR could further expand the capabilities of managing vast amounts of glucose data in real-time to predict trends, detect abnormalities, and direct insulin delivery in a more precise manner. Robust research studies are needed to examine the multiple facets required to successfully and seamlessly integrate CGMs into the perioperative setting. Firstly, interoperability standards should be developed and refined by using existing frameworks such as the FHIR in order to ensure smooth data exchange. Secondly, methods to preserve data privacy including secure data transfer, encryption methods, and compliance with HIPPA regulations are essential. Thirdly, AI algorithm integration should be explored to analyze CGM data within the EHR. Lastly, further research regarding benefits of the integration of CGM data into the health record could create workflows to improve OR efficiency and perioperative patient outcomes. This could be accomplished with prospective clinical trials. Additionally, successful integration of CGMs will heavily depend on key stakeholders (healthcare providers, healthcare systems, device manufacturers, information technology departments and administrative bodies) who must collaborate to develop evidence based standardized perioperative guidelines for diabetic patients. Proper patient selection, comprehensive education, adherence to applicable guidelines, and a seamless integration with hospital information systems will allow CGMs to revolutionize perioperative pediatric diabetic care, thereby enhancing patient quality of life.

ACKNOWLEDGEMENTS

I would like to extend my heartfelt appreciation to my young friend, whose journey with TID has been nothing short of inspiring. Through courage and resilience, he has successfully managed this life-changing condition with the aid of a CGM. I would also like to express my sincere gratitude to the faculty members of the Department of Anesthesiology at Westchester Medical Center for their support and encouragement of academic endeavors.